covid economics · 2020. 12. 18. · the economics of electoral politics and small business loans...
TRANSCRIPT
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
FROM THE GREAT RECESSION TO THE PANDEMIC RECESSIONFrancis X. Diebold
ELECTORAL POLITICS AND SMALL BUSINESS LOANSRan Duchin and John Hackney
GROWTH FORECASTS AT END-2020Javier G. Gómez-Pineda
STOP-AND-GO EPIDEMIC CONTROLClaudius Gros and Daniel Gros
CONSUMPTION RESPONSES TO STIMULUS PAYMENTSSo Kubota, Koichiro Onishi and Yuta Toyama
CHILD CARE CLOSURES AND WOMEN’S WORKLauren Russell and Chuxuan Sun
GRANDPARENTAL CHILD CAREChristina Boll and Till Nikolka
ISSUE 62 18 DECEMBER 2020
Covid Economics Vetted and Real-Time PapersCovid Economics, Vetted and Real-Time Papers, from CEPR, brings together formal investigations on the economic issues emanating from the Covid outbreak, based on explicit theory and/or empirical evidence, to improve the knowledge base.
Founder: Beatrice Weder di Mauro, President of CEPREditor: Charles Wyplosz, Graduate Institute Geneva and CEPR
Contact: Submissions should be made at https://portal.cepr.org/call-papers-covid-economics. Other queries should be sent to [email protected].
Copyright for the papers appearing in this issue of Covid Economics: Vetted and Real-Time Papers is held by the individual authors.
The Centre for Economic Policy Research (CEPR)
The Centre for Economic Policy Research (CEPR) is a network of over 1,500 research economists based mostly in European universities. The Centre’s goal is twofold: to promote world-class research, and to get the policy-relevant results into the hands of key decision-makers. CEPR’s guiding principle is ‘Research excellence with policy relevance’. A registered charity since it was founded in 1983, CEPR is independent of all public and private interest groups. It takes no institutional stand on economic policy matters and its core funding comes from its Institutional Members and sales of publications. Because it draws on such a large network of researchers, its output reflects a broad spectrum of individual viewpoints as well as perspectives drawn from civil society. CEPR research may include views on policy, but the Trustees of the Centre do not give prior review to its publications. The opinions expressed in this report are those of the authors and not those of CEPR.
Chair of the Board Sir Charlie BeanFounder and Honorary President Richard PortesPresident Beatrice Weder di MauroVice Presidents Maristella Bott cini Ugo Panizza Philippe Martin Hélène ReyChief Executive Officer Tessa Ogden
Editorial BoardBeatrice Weder di Mauro, CEPRCharles Wyplosz, Graduate Institute Geneva and CEPRViral V. Acharya, Stern School of Business, NYU and CEPRGuido Alfani, Bocconi University and CEPRFranklin Allen, Imperial College Business School and CEPRMichele Belot, Cornell University and CEPRDavid Bloom, Harvard T.H. Chan School of Public HealthTito Boeri, Bocconi University and CEPRAlison Booth, University of Essex and CEPRMarkus K Brunnermeier, Princeton University and CEPRMichael C Burda, Humboldt Universitaet zu Berlin and CEPRAline Bütikofer, Norwegian School of EconomicsLuis Cabral, New York University and CEPRPaola Conconi, ECARES, Universite Libre de Bruxelles and CEPRGiancarlo Corsetti, University of Cambridge and CEPRFiorella De Fiore, Bank for International Settlements and CEPRMathias Dewatripont, ECARES, Universite Libre de Bruxelles and CEPRJonathan Dingel, University of Chicago Booth School and CEPRBarry Eichengreen, University of California, Berkeley and CEPRSimon J Evenett, University of St Gallen and CEPRMaryam Farboodi, MIT and CEPRAntonio Fatás, INSEAD Singapore and CEPRPierre-Yves Geoffard, Paris School of Economics and CEPRFrancesco Giavazzi, Bocconi University and CEPRChristian Gollier, Toulouse School of Economics and CEPRTimothy J. Hatton, University of Essex and CEPREthan Ilzetzki, London School of Economics and CEPRBeata Javorcik, EBRD and CEPRSimon Johnson, MIT and CEPR
Sebnem Kalemli-Ozcan, University of Maryland and CEPR Rik FrehenTom Kompas, University of Melbourne and CEBRAMiklós Koren, Central European University and CEPRAnton Korinek, University of Virginia and CEPRMichael Kuhn, Vienna Institute of DemographyMaarten Lindeboom, Vrije Universiteit AmsterdamPhilippe Martin, Sciences Po and CEPRWarwick McKibbin, ANU College of Asia and the PacificKevin Hjortshøj O’Rourke, NYU Abu Dhabi and CEPREvi Pappa, European University Institute and CEPRBarbara Petrongolo, Queen Mary University, London, LSE and CEPRRichard Portes, London Business School and CEPRCarol Propper, Imperial College London and CEPRLucrezia Reichlin, London Business School and CEPRRicardo Reis, London School of Economics and CEPRHélène Rey, London Business School and CEPRDominic Rohner, University of Lausanne and CEPRPaola Sapienza, Northwestern University and CEPRMoritz Schularick, University of Bonn and CEPRPaul Seabright, Toulouse School of Economics and CEPRFlavio Toxvaerd, University of CambridgeChristoph Trebesch, Christian-Albrechts-Universitaet zu Kiel and CEPRKaren-Helene Ulltveit-Moe, University of Oslo and CEPRJan C. van Ours, Erasmus University Rotterdam and CEPRThierry Verdier, Paris School of Economics and CEPR
EthicsCovid Economics will feature high quality analyses of economic aspects of the health crisis. However, the pandemic also raises a number of complex ethical issues. Economists tend to think about trade-offs, in this case lives vs. costs, patient selection at a time of scarcity, and more. In the spirit of academic freedom, neither the Editors of Covid Economics nor CEPR take a stand on these issues and therefore do not bear any responsibility for views expressed in the articles.
Submission to professional journalsThe following journals have indicated that they will accept submissions of papers featured in Covid Economics because they are working papers. Most expect revised versions. This list will be updated regularly.
American Economic Review American Economic Review, Applied EconomicsAmerican Economic Review, InsightsAmerican Economic Review, Economic Policy American Economic Review, Macroeconomics American Economic Review, Microeconomics American Journal of Health EconomicsCanadian Journal of EconomicsEconometrica*Economic JournalEconomics of Disasters and Climate ChangeInternational Economic ReviewJournal of Development EconomicsJournal of Econometrics*
Journal of Economic GrowthJournal of Economic TheoryJournal of the European Economic Association*Journal of FinanceJournal of Financial EconomicsJournal of International EconomicsJournal of Labor Economics*Journal of Monetary EconomicsJournal of Public EconomicsJournal of Public Finance and Public ChoiceJournal of Political EconomyJournal of Population EconomicsQuarterly Journal of EconomicsReview of Corporate Finance Studies*Review of Economics and StatisticsReview of Economic Studies*Review of Financial Studies
(*) Must be a significantly revised and extended version of the paper featured in Covid Economics.
Covid Economics Vetted and Real-Time Papers
Issue 62, 18 December 2020
Contents
Real-time real economic activity entering the pandemic recession 1Francis X. Diebold
Buying the vote? The economics of electoral politics and small business loans 20Ran Duchin and John Hackney
Growth forecasts and the Covid-19 recession they convey: End-2020 update 66Javier G. Gómez-Pineda
The economics of stop-and-go epidemic control 74Claudius Gros and Daniel Gros
Consumption responses to COVID-19 payments: Evidence from a natural experiment and bank account data 90So Kubota, Koichiro Onishi and Yuta Toyama
The effect of mandatory child care center closures on women’s labor market outcomes during the COVID-19 pandemic 124Lauren Russell and Chuxuan Sun
Rightly blamed the ‘bad guy’? Grandparental child care and Covid-19 155Christina Boll and Till Nikolka
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
Covid Economics Issue 62, 18 December 2020
Copyright: Francis X. Diebold
Real-time real economic activity entering the pandemic recession1
Francis X. Diebold2
Date submitted: 15 December 2020; Date accepted: 16 December 2020
Entering the Pandemic Recession, we study the high-frequency real-activity signals provided by a leading nowcast, the ADS Index of Business Conditions produced and released in real time by the Federal Reserve Bank of Philadelphia. We track the evolution of real-time vintage beliefs and compare them to a later-vintage chronology. Real-time ADS plunges and then swings as its underlying economic indicators swing, but the ADS paths quickly converge to indicate a return to brisk positive growth by mid-May. We show, moreover, that the daily real activity path was extremely highly correlated with the daily COVID-19 path. Finally, we provide a comparative assessment of the real-time ADS signals provided when exiting the Great Recession.
1 For helpful discussion I thank Boragan Aruoba, Scott Brave, Andrew Patton, Glenn Rudebusch, Frank Schorfh ide, Chiara Scotti, Minchul Shin, Keith Sill, Mark Watson, Tom Stark, Jim Stock, Herman van Dijk,and Simon van Norden. I am also grateful to conference/seminar participants at the Federal Reserve Bank of Philadelphia, the Society for Financial Econometrics, Amazon, the Department of Statistics at the Wharton School, and the International Association for Applied Econometrics. For outstanding research assistance and related discussion I thank Philippe Goulet Coulombe, Tony Liu, and Boyan Zhang. The usual disclaimer applies.
2 Paul F. Miller, Jr. and E. Warren Shafer Miller Professor of Social Sciences, and Professor of Economics, Finance, and Statistics, University of Pennsylvania.
1
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 1-
19
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
1 Introduction
Accurate assessment of of current real economic activity (“business conditions”) is key for
successful decision making in business, finance, and policy. It is difficult, however, to track
business conditions in real time, both because no single observed economic indicator is
“business conditions”, and because different indicators are available at different observa-
tional frequencies, and with different release delays. Nevertheless there exists the tantalizing
possibility of accurate real-time business conditions assessment (“nowcasting”), and recent
decades have witnessed great interest in nowcasting methods and applications (e.g., Banbura
et al. (2011)).
The workhorse nowcasting approaches involve dynamic factor models, which relate a
set of observed real activity indicators to a single underlying latent real activity factor.
Both “small data” approaches (e.g., based on 5 indicators) and “Big Data” approaches
(e.g., based on 500 indicators) are available. Small data approaches were developed first,
and they typically involve maximum likelihood estimation (e.g., Stock and Watson (1989)).
Subsequent Big Data approaches, in contrast, typically involve two-step estimation based on
a first-step extraction of principal components (e.g., Stock and Watson (2002), McCracken
and Ng (2016)).
Both introspection and experience reveal that Big Data nowcasting approaches are not
necessarily better. First, they are more tedious to manage, and less transparent. Second,
they may not deliver much improvement in factor extraction accuracy, which increases and
stabilizes quickly as the number of indicators increases (Doz et al., 2012). Third, casual
inclusion of many indicators can be problematic because a poorly-balanced set of indicators
can create distortions in the extracted factor (Boivin and Ng, 2006), whereas small data
approaches promote and facilitate hard thinking about a well-balanced set of indicators (Bai
and Ng (2008)).
Against this background, in this paper we assess the performance of a leading small-data
nowcast, the Aruoba-Diebold-Scotti (ADS) Index of Business Conditions (Aruoba et al.,
2009). ADS is designed to track real business conditions at high frequency, and it has been
maintained and released in real time by the Federal Reserve Bank of Philadelphia contin-
uously since 2008.1 Its modeling style and underlying economic indicators build on classic
1The production version used by FRB Philadelphia differs in some ways (e.g., included indicators andtreatment of trend) from the prototypes provided by Aruoba et al. (2009) and Aruoba and Diebold (2010),which themselves differ slightly. All discussion in this paper refers to the FRB Philadelphia version. Allmaterials, including the full set of vintage nowcasts, are available at https://www.philadelphiafed.org/
research-and-data/real-time-center/business-conditions-index.
2
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 1-
19
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
early work in the tradition of Burns and Mitchell (1946), Sargent and Sims (1977), and Stock
and Watson (1989). The underlying indicators span high- and low-frequency information on
real economic flows: weekly initial jobless claims; monthly payroll employment growth, in-
dustrial production growth, personal income less transfer payments growth, manufacturing
and trade sales growth; and quarterly real GDP growth.
Crucially, we assess ADS using only information actually available in real time. This
is required for truly credible real-time evaluation, and it can only be achieved by using
nowcasts produced and permanently recorded in real time, which is very different from simply
removing final-revised data and inserting vintage data into an otherwise ex post analysis.
Unfortunately, such evaluations are rare, because there simply are not many instances of long
series of nowcasts produced and recorded in real time. ADS, however, has been produced
and recorded in real time roughly twice weekly since late 2008, so we can provide real-
time performance assessments both exiting the Great Recession and entering the Pandemic
Recession.
We proceed as follows. In section 2 we provide background on aspects of ADS construc-
tion, updating, ex post characteristics, and performance evaluation. In section 3 we examine
ADS entering the Pandemic Recession, and we relate the real-time ADS path to the real-
time COVID-19 path. In section 4 we provide a comparative examination of ADS exiting
the Great Recession. We conclude in section 5.
2 Nowcast Construction, Characteristics, and Assess-
ment
Here we provide background on the ADS index construction (section 2.1), ex post historical
characteristics (section 2.2), and general issues of relevance to assessing ex ante nowcasting
performance (section 2.3).
2.1 Construction and Updating
ADS is a dynamic factor model with multiple mixed-frequency real activity indicators driven
by a single latent real activity factor. The ADS index is an estimate of that latent real activity
factor. Importantly, the model is specified such that the real activity factor tracks the de-
meaned growth rate of real activity. Progressively more negative or positive values indicate
progressively worse- or better-than-average real growth, respectively. Because ADS tracks
3
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 1-
19
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
real activity growth, not level, a positive value does not necessarily mean “good times”;
rather, it means “good growth”, which may be from a level well below trend, as for example
in the early stages of a recovery.
ADS is specified at daily frequency, allowing as necessary for missing data for the less-
frequently observed variables.2 Importantly, despite complications from missing data, time-
varying system matrices, aggregation across frequencies, etc., the Kalman filter and associ-
ated Gaussian pseudo likelihood evaluation via prediction-error decomposition remain valid,
subject to some well-known modifications.3 Model estimation is therefore straightforward,
after which the Kalman smoother produces an optimal extraction of the underlying real
activity factor. That is, the Kalman smoother produces the ADS index: The extracted
sequence at any time t∗ is the vintage-t∗ ADS sequence, {ADS1, ADS2, ...ADSt∗}.
The first ADS vintage was released 12/5/2008, covering 3/1/1960 through 11/30/2008.
Since then, ADS has been continuously updated whenever new data are released. The
Kalman smoother is re-run, generally within two hours of the release, and the newly-
extracted index from 3/1/1960 to “the present” is re-written to the web. ADS has been
updated approximately eight times per month on average since inception.
2.2 Ex Post Characteristics
In Figure 1 we show the ADS index from 03/01/1960 through 12/31/2013, as assessed in the
6/26/2020 vintage. The sample range is well before the vintage pull date, so the chronology
displayed is (intentionally) ex post. We do this because it is instructive to examine the ex
post chronology before passing to real time assessment, which can only be done after ADS
went live in late 2008.4
Several features are noteworthy. For example, the ADS chronology coheres strongly with
the NBER chronology, plunging during NBER recessions. In addition, several often-discussed
features of the business-cycle are evident in ADS, such as the pronounced moderation in
volatility during the Greenspan era.
The ADS value added relative to the NBER chronology stems from the facts that (1) it is
a cardinal measure, allowing one to assess not only recession durations, but also depths and
2The model must be specified at daily frequency, despite the fact that the highest-frequency indicatoris is weekly initial jobless claims, to account for the varying number of days/weeks per month, which alsoproduces time-varying system parameter matrices.
3See, for example, Durbin and Koopman (2001) on missing data, and Harvey (1991) on aggregation offlow variables.
4The sample period intentionally excludes the Pandemic Recession, which we will subsequently examinein detail.
4
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 1-
19
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
Figure 1: ADS Index: Ex Post Path 03/01/1960 - 12/31/2013 (Vintage 6/26/2020)
Notes: The shaded regions are NBER-designated recessions.
patterns (see Table 1), and (2) its updates arrive in timely fashion, whereas the starting and
ending dates NBER recessions are typically not announced until well after the fact (again
see Table 1). Of course, if ADS is to be a useful guide for business and policy decisions, its
frequently-arriving updates must provide reliable signals in real time, not just ex post as in
Figure 1. We now turn to that issue.
2.3 Performance Assessment
Truly credible nowcasting performance assessment requires using vintage information, which
emerges as the limit of a sequence of progressively more realistic and credible nowcast/forecast
evaluation approaches:5
(1) Use full-sample estimation, and use final revised data
(2) Use expanding-sample estimation, and use final revised data
(3) Use expanding-sample estimation, and use vintage data (“Pseudo Real Time”)
(4) Use expanding-sample estimation, and use vintage information (“Real Time”).
5Note that nowcasts are effectively just h-step-ahead forecasts with horizon h=0.
5
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 1-
19
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
Table 1: NBER Recessions
Recession Dates Recession Characteristics
Peak Month Trough Month Duration Depth Severity
April 1960 February 1961 10 2.7 27.0December 1969 November 1970 11 2.8 30.8November 1973 March 1975 16 4.7 75.2January 1980 (6/3/1980) July 1980 (7/8/1981) 6 3.6 21.6July 1981 (1/6/1982) November 1982 (7/8/1983) 16 2.9 46.4July 1990 (4/25/1991) March 1991 (12/22/1992) 8 1.7 13.6March 2001 (11/26/2001) November 2001 (7/17/2003) 8 1.5 12.0December 2007 (12/1/2008) June 2009 (9/20/2010) 18 4.3 77.4February 2020 (6/6/2020) ? ? ? ?
Notes: Recession dates and durations in months are from the NBER chronology; see https://www.nber.
org/cycles.html. When available, the announcement dates appear in parentheses. The NBER troughmonth for the Pandemic Recession has not yet been announced. Recession depth is the minimum absolutedaily ADS value during the recession; more precisely, the depth D of recession R is D = |mini(ADSi)|,i ∈ R, where i denotes days. Recession severity S is the product of depth and duration. Both D and S usea late-vintage ADS chronology and the NBER recession chronology.
Approaches (1) and (2) are clearly unsatisfactory: Approach (1) uses time periods and
data values not available in real time, and approach (2) is an improvement but still uses
data values not available in real time. Approach (3), involving vintage data, is typically
viewed as the gold standard. It is implemented comparatively infrequently, however, due to
the tedium involved and the fact that vintage data are often unavailable.6 Approach (4),
involving vintage information, limits the information set to that available and actually used
in real time, which is more restrictive than merely limiting the data to that available in real
time. It is, however, almost never implemented.
To appreciate why fully-credible assessment requires vintage information rather than just
vintage data, consider the following:
(1) Econometric/statistical theory and experience evolve, prompting changes to the esti-
mation procedure; the frequency and timing of re-estimation and its interaction with
benchmark revisions; the estimation sample period; allowance for parameter variation
6The two key sources of U.S. vintage data are the Real-Time Dataset for Macroeconomists atthe Federal Reserve Bank of Philadelphia (https://www.philadelphiafed.org/research-and-data/real-time-center/real-time-data/), and ALFRED at the Federal Reserve Bank of St. Louis (https://alfred.stlouisfed.org/).
6
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 1-
19
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
and breaks; the treatment of outliers; the strength of regularization employed; the
predictive loss function employed; etc.
(2) Economic theory and empirical economic experience evolve. Over time this may
prompt, for example, the removal or re-weighting of some component nowcast indi-
cators and/or addition of others (e.g., Diebold and Rudebusch (1991)), as well as
deeper changes in the nowcasting model.
(3) Exact times and reliability of nowcast/forecast calculation and release may differ due
to technological problems; outright mistakes in nowcast/forecast construction; evolving
or changing software algorithms and associated bugs; parallel problems at the agencies
responsible for the underlying data and decisions regarding how to deal with them in
forecast/nowcast construction; etc.
For these and other reasons, just as truly credible evaluation requires refraining from en-
dowing agents with better data than were actually available in real time, so too does it
require refraining from endowing them with better economic or statistical models and re-
lated tools than were actually available in real time, better judgment and decision-making
abilities/choices than were actually manifest in real time, etc.
The upshot is clear: Truly credible real-time evaluation – that is, evaluation using vintage
information rather than just vintage data – can only be obtained by using nowcasts produced
and permanently recorded in real time. ADS has been produced and recorded in real time
since late 2008, so we can credibly study the key episode of current interest, entry into the
Pandemic Recession. We now proceed to do so.
3 The Pandemic Recession Entry
We focus in this section on the Pandemic Recession that started in March 2020. It is
instructive to begin by comparing it to the Great Recession of 2007-2009. To that end we
show the ADS path in Figure 2, from late 2007 through June 2020.7 The so-called “Great
Recession” appears minor by comparison.
7We refer to an ADS extraction as a path.
7
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 1-
19
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
Figure 2: ADS Index: Ex Post Path 12/1/2007 - 6/26/2020 (Vintage 6/26/2020)
3.1 A Detailed Look at the Later-Vintage Path
Figure 2 reveals the jaw-dropping ADS drop in the Pandemic Recession, more than five times
that of any other recession since 1960. The ADS drop is entirely appropriate, due to similarly
jaw-dropping and historically unprecedented movements in its underlying indicators.
As of this writing, the official trough month for the Pandemic Recession has not been
announced. It could be as early as May 2020, in which case the Pandemic Recession would
be the shortest in history. Indeed a May trough turns out to be likely. In Figure 3 we
show the later-vintage Pandemic Recession path. The overall extracted path is smooth and
convex, with a minimum in early April, and a return to positive growth by mid-May. We
emphasize again, however, that ADS measures real activity growth, not level. Hence positive
ADS does not necessarily mean “good times”; rather, it means “good growth”, which may
be from a very bad initial condition. That was the situation in late May, as the battered
U.S. economy evidently resumed growth.
3.2 Real-Time Vintages
3.2.1 Five Snapshots
In Figure 4 we show several end-of-month paths in black, starting with February 2020. For
comparison, in each panel we also show the later-vintage path in red. Moving through the
five panels of Figure 4:
8
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 1-
19
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
Figure 3: ADS Index: Ex Post Path 1/1/2020 - 6/26/2020 (Vintage 6/26/2020)
(1) In the top panel we show the 2/28/2020 path. ADS has not moved.
(2) In the second panel we show the 3/27/2020 path, which looks very different. ADS has
become acutely aware of the disastrous situation; indeed most of the 3/27 path is well
below the previous all-time (post-1960) ADS low during the 1970s oil-shock recession.8
(3) In the third panel we show the 4/30/2020 path. The April initial claims news is bad,
but less bad than March, which is good, and ADS shows a minimum in late March
followed by a rise toward normalcy by the end of April.
(4) In the fourth panel we show the 5/29/2020 path. The May news is very bad, dominated
by the shockingly bad May 8 payroll employment number (for April), and the late-May
path is massively down-shifted relative to the late-April path. The new minimum is
in mid-April rather than late March, and the 5/29 ADS value is thoroughly dismal,
nowhere near normalcy.
(5) In the fifth panel we show the 6/26/2020 path. Thanks to the strong May payroll
employment number (released June 5), ADS moved into normal territory, and stayed
there. There is clear (albeit highly-tentative evidence for a Pandemic Recession trough
in mid-May, when ADS hits 0.)
8It is also apparent that the Kalman smoother may be smoothing “too much”, producing low ADS valueswell before mid-March, going back into February and even January. Its smoothing is optimal relative to thepatterns in historical data, but the March initial jobless claims movements were unprecedentedly sharp.
9
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 1-
19
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
Figure 4: Entering the Pandemic Recession: Monthly Real-Time ADS Paths
Notes: We show five monthly real-time ADS paths in black. From top to bottom they are2/28/2020, 3/27/2020, 4/30/2020, 5/29/2020, and 6/26/2020. For comparison we show the6/26/2020 path in red in all panels. (In the bottom panel, we show only black, since blackand red are identical.)
10
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 1-
19
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
Figure 5: Entering the Pandemic Recession: Real Time ADS Path Plot
Notes: We show all real-time ADS paths in black, through 6/26/2020. For comparison weshow the complete later-vintage path (6/26/2020) in red.
3.2.2 The Full Path Plot and Dot Plot
In Figure 5 we show the complete path plot during the Pandemic Recession through 6/26/2020,
with the later-vintage path in red for comparison. The path plot is the set of all real-time
paths; by following rightward through the sequence of paths, moving through time, we track
the evolution of ADS beliefs about the chronology of business conditions.
There are wide real-time divergences between individual early paths and the later vintage
red path. There are interesting patterns, however, with several real-time “meta paths”
evident:
(1) The first extends through the 3/19/2020 ADS announcement. ADS does not move.
Initial claims rise from 0.2m to 0.3m, a large move by historical standards, confirm-
ing what everyone already knew: the pandemic would have important real economic
consequences, but the Kalman smoother optimally but erroneously ascribes it to mea-
surement error.
(2) The second meta-path begins with the 3/26/2020 and 4/2/2020 initial claims explo-
sions. ADS plunges, but then recovers steadily despite a steady stream of bad news
11
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 1-
19
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
Figure 6: Entering the Pandemic Recession: Real-Time ADS Dot Plot
Notes: We show the last values of all real-time ADS paths in black. For comparison we showthe complete later-vintage path (6/26/2020) in red.
(it is bad, but getting less bad), almost back to 0 by the 5/7/2020 initial claims an-
nouncement.
(3) The third meta-path begins with the horrific 5/8/2020 April payroll employment re-
lease, with ADS again plunging. It then again begins mean reverting, and does so
completely when the strong May payroll employment number is released on 6/5/2020.
In Figure 6 we show the corresponding “dot plot”, with the 6/26/2020 path again superim-
posed. Each dot is the last observation of its corresponding path in Figure 5. The dots are
real-time filtered values, because smoothed and filtered values coincide for the last observa-
tion in a sample. The dot plot is highly volatile and emphasizes the various meta-paths.
3.3 Real Economic Activity and COVID-19
Because the March-April 2020 collapse in economic activity was obviously caused by COVID-
19, it is of interest to directly examine the correlation between the two. We can do so at high
frequency (daily), because we have both daily ADS and COVID new cases / deaths data. We
want to correlate COVID new cases with ADS, but the direct new cases data are less reliable
12
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 1-
19
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
Figure 7: Daily ADS and Smoothed Daily COVID-19 Deaths
Notes: We show ADS (6/26/2020 vintage) vs HP-filtered daily COVID-19 deaths led by 20days. See text for details.
than deaths during the period of interest, because new cases were likely heavily influenced
by changes in the amount of testing undertaken. Instead, a more reliable indicator of new
cases is deaths, adjusted for the approximate 20-day period between infection and death.
Hence we use deaths led by 20 days.9 In Figure 7 we show ADS vs COVID deaths+20.10
The strength of the negative correlation is striking. Of course economic activity plunged in
March when COVID exploded, but there’s much more than that – ADS and COVID continue
to move in lockstep (inversely) through the April COVID peak, its April-May decline, and
its June rebound.
4 Comparison to the Great Recession Exit
It is informative to compare the evolution and congealing of views during the Pandemic
Recession entry to those during an earlier, more “standard”, recession, like the Great Re-
9We use the Johns Hopkins University CSSE COVID-19 daily deaths data; see Dong et al. (2020) andhttps://github.com/CSSEGISandData/COVID-19.
10We also smooth COVID deaths+20 using a Hodrick-Prescott filter to remove the strong calendar effectsin recorded deaths.
13
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 1-
19
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
Figure 8: Exiting the Great Recession: Five Quarterly Real-Time ADS Paths
Notes: We show five quarterly real-time ADS paths in black. From top to bottom theyare 12/5/2008, 3/6/2009, 6/5/2009, 9/3/2009, 12/4/2009. For comparison we show a later-vintage ADS path (December 2010) in red.
14
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 1-
19
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
cession of 2007-2009. We can’t examine real-time ADS when entering the Great Recession,
because ADS did not start until December 2008, well after the great recession began. But
we can examine it when exiting the great recession. In Figure 8 we show five paths in black,
from ADS inception through the end of the Great Recession, at quarterly intervals. For
comparison we also show a later-vintage path in red.
In the top panel of Figure 8 we show the first ADS path, 12/5/2008. ADS shows a very
deep recession, almost the deepest on record since 1960, bottoming out in 2008Q3, with
movement toward recovery in late Q3 and early Q4, even if it had stalled a bit by early
December. As it turned out, however, the Great Recession subsequently featured a growth
rate “double dip”. The 12/5/2008 ADS path ends just after the first dip, which involved a
sharp drop in September 2008 and an equally sharp rebound.11 At the time it was easy to
read the cards as saying that the recession was ending, and ADS was a bit too optimistic,
moving upward toward recovery.
Now consider the remaining panels of Figure 8. In the second panel we show the next,
and contrasting, 3/6/2009 ADS path. In the interim ADS has quickly learned the situation,
the double dip in particular, and is very much on track, capturing the second dip in January
2009. ADS continues to climb steadily through the third and fourth panels (6/5/2009 and
9/3/3009, respectively), and by the time of the bottom panel (12/4/2009) it is clear that the
Great Recession ended in June or July, with ADS basically fluctuating around 0 after that.
(Recall that ADS=0 means average growth, not zero growth.)
All told, the five quarterly real-time ADS paths generally match the ex post path closely,
and they correctly identify the recession’s end, well before the end of 2009 and indeed roughly
1.5 years before the official NBER announcement in September 2010.
To emphasize ADS timeliness, we plot the later-vintage ADS in Figure 8 all the way
through 2010, which allows inclusion of the NBER’s end-of-recession announcement on
9/20/2010, long after the fact and not helpful for real-time decision making.12 ADS fills
the gap left by the late-arriving NBER chronology, and it also provides a numerical measure
that allows one to track the recession’s pattern, depth, overall severity, etc., in addition to
11In particular, according to the Federal Reserve’s G.17 Industrial Production (IP) release of October16, 2008, September IP was severely affected by a highly-unusual and largely exogenous “triple shock”(Hurricanes Gustav and Ike, and a strike at a major aircraft manufacturer), which caused an annualizedSeptember IP drop of nearly fifty percent. A similar pattern exists for Manufacturing and Trade Sales(MTS). IP and MTS also rebounded unusually sharply in October – indeed IP appears to “overshoot” –presumably in an attempt by manufacturers to make up for September’s loss.
12Of course the NBER is not seeking to be helpful for real-time decision making; rather, they seek tometiculously construct the U.S. business cycle chronology of record, quite reasonably using all relevantinformation – even very late-arriving information.
15
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 1-
19
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
Figure 9: Exiting the Great Recession: Real-Time ADS Path Plot
Notes: We show all 2008-2009 ADS paths since the first on 12/5/2008. We show real-timeADS paths in black, and a comparison late-vintage ADS path (December 2010) in red.
Figure 10: Exiting the Great Recession: Real-Time ADS Dot Plot
Notes: We show the last values of each 2008-2009 ADS path in black, with a comparisonlater-vintage ADS path (December 2010) superimposed in red.
16
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 1-
19
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
duration. For example and as recorded in Table 1, ADS identifies the Great Recession as
the worst since 1960 and through 2010, with longest duration and second-greatest depth,
resulting in the greatest overall severity (duration times depth).
In Figure 9 we show the complete path plot. Of course there are errors positive and
negative as the recession evolves, but overall ADS performs well, sending a reliable and
valuable signal for navigating the path out of recession. We show the corresponding dot plot
in Figure 10.
5 Conclusion
We explored how views formed using a leading nowcast (ADS) evolved when entering the
U.S. Pandemic Recession, which arrived abruptly and was caused by non-economic factors,
tracking the evolution of real-time vintage beliefs and comparing them to a later-vintage
chronology. ADS real activity growth plunged wildly in March 2020 and swung in real
time as its underlying components swung, but it clearly returned to brisk growth by mid
May, making the Pandemic Recession surely the deepest and likely the shortest on record.13
We also documented a strong negative relationship between the real-time ADS Pandemic
Recession entry path and the concurrent real-time COVID-19 entry path, and we compared
the ADS Pandemic Recession entry path to the earlier Great Recession exit path.
13The NBER has not yet announced the ending date of the Pandemic Recession.
17
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 1-
19
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
References
Aruoba, S.B. and F.X. Diebold (2010), “Real-Time Macroeconomic Monitoring: Real Ac-
tivity, Inflation, and Interactions,” American Economic Review , 100, 20–24.
Aruoba, S.B., F.X. Diebold, and C. Scotti (2009), “Real-time Measurement of Business
Conditions,” Journal of Business and Economic Statistics , 27, 417–427.
Bai, J. and S. Ng (2008), “Forecasting Economic Time Series Using Targeted Predictors,”
Journal of Econometrics , 146, 304–317.
Banbura, M., D. Giannone, and L. Reichlin (2011), “Nowcasting,” In M. Clemens and D.
Hendry, eds., Oxford Handbook of Economic Forecasting, 193-224.
Boivin, J. and S. Ng (2006), “Are More Data Always Better for Factor Analysis?” Journal
of Econometrics , 132, 169–194.
Burns, A. and W.C. Mitchell (1946), Measuring Business Cycles, Columbia University Press
for NBER.
Diebold, F.X. and G.D. Rudebusch (1991), “Forecasting Output with the Composite Leading
Index: A Real-Time Analysis,” Journal of the American Statistical Association, 86, 603–
610.
Dong, E., H. Du, and L. Gardner (2020), “An Interactive Web-Based Dashboard to Track
COVID-19 in Real Time,” The Lancet Infectious Diseases , 20, 533–534.
Doz, C., D. Giannone, and L. Reichlin (2012), “A Quasi–Maximum Likelihood Approach for
Large, Approximate Dynamic Factor Models,” Review of Economics and Statistics , 94,
1014–1024.
Durbin, J. and S.J. Koopman (2001), Time Series Analysis by State Space Methods , Oxford:
Oxford University Press.
Harvey, A.C. (1991), Forecasting, Structural Time Series Models and the Kalman Filter ,
Cambridge: Cambridge University Press.
McCracken, M. and S. Ng (2016), “FRED-MD: A Monthly Database for Macroeconomic
Research,” Journal of Business and Economic Statistics , 36, 574–589.
18
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 1-
19
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
Sargent, T.J. and C.A. Sims (1977), “Business Cycle Modeling Without Pretending to Have
too Much a Priori Theory,” In C.A. Sims (ed.), New Methods in Business Cycle Research:
Proceedings from a Conference, Federal Reserve Bank of Minneapolis, 45-109.
Stock, J. H. and M.W. Watson (1989), “New Indexes of Coincident and Leading Economic
Indicators,” In O. J. Blanchard and S. Fischer, eds., NBER Macroeconomics Annual, MIT
Press, 351-393.
Stock, J. H. and M.W. Watson (2002), “Forecasting Using Principal Components from a
Large Number of Predictors,” Journal of the American Statistical Association, 97, 147–
162.
19
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 1-
19
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
Covid Economics Issue 62, 18 December 2020
Copyright: Ran Duchin and John Hackney
Buying the vote? The economics of electoral politics and small business loans1
Ran Duchin2 and John Hackney3
Date submitted: 14 December 2020; Date accepted: 15 December 2020
We study the role of electoral politics in government small business lending, employment, and business formation. We construct novel measures of electoral importance capturing swing and base voters using data from Facebook ad spending, independent political expenditures, the Cook Political Report, and campaign contributions. We find that businesses in electorally important states, districts, and sectors receive more loans following the onset of the Covid-19 crisis, controlling for funding demand and both health and economic conditions. Estimates from survey and observational data show that government funding weakens the adverse effects of the crisis on employment, small business activity, and business applications.
1 We thank conference participants at the 2020 International Conference of Taiwan Finance Association.2 Carroll School of Management, Boston College.3 Darla Moore School of Business, University of South Carolina.
20
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 2
0-65
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
1. Introduction
In this paper, we seek to provide novel empirical evidence on the role of election-year political
incentives in the government’s allocation of emergency funds and their real economic effects. We
focus our attention on the Covid-19 outbreak, which was an unexpected economy-wide shock that
triggered a large-scale government aid response. This response disbursed trillions of dollars across
states, businesses, and individuals during a period of economic stress, when the benefit of
government aid is potentially greatest. The outbreak also coincides with the 2020 presidential
election year in the U.S, which is characterized by strident political polarization. According to
Gallup, 82 percentage points separate Republicans’ (89%) and Democrats’ (7%) average job
approval ratings of President Trump during his third year in office -- the largest degree of political
polarization in any presidential year measured by Gallup.1
We argue that the confluence of a massive emergency government aid package and a
polarized presidential race generates a unique setting to identify the role of electoral politics in the
allocation of government funds and its economic consequences. In particular, a large body of
evidence in political economy suggests that voters reward incumbents based on economic
conditions in the year before Election Day rather than throughout their tenure (e.g., Kramer 1971;
Fair 1978; Kiewiet 1983; Alesina, Londregan, and Rosenthal 1993; Achen and Bartels 2004).
Furthermore, Achen and Bartels (2004) conclude that long-term economic growth contributes little
or nothing to the incumbent party’s electoral prospects. Such voter behavior introduces incentives
to implement election-year policies that improve the reelection prospects of incumbents, possibly
1 See: “Trump Third Year Sets New Standard for Party Polarization,” by Jeffrey M. Jones, https://news.gallup.com/poll/283910/trump-third-year-sets-new-standard-party-polarization.aspx
21
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 2
0-65
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
with considerable economic effects and at the cost of long-term economic growth (e.g., Tufte
1978).
The empirical analyses focus on the Paycheck Protection Program (PPP), which is a central
piece of the 2020 Coronavirus Aid, Relief, and Economic Security (CARES) Act. The PPP was
administered by the Small Business Administration (SBA) and extended forgivable loans to
businesses to cover payroll, utilities, mortgage, and rent costs. The combination of the attractive
terms of the PPP and the sharp decline in economic activity resulting from the shelter-in-place
policies implemented in response to Covid-19 led to oversubscription to the PPP and consequently
to credit rationing. As such, the PPP could have been a powerful instrument to implement election-
year allocative policies. Using detailed data on the allocation of forgivable PPP loans, we
investigate how the politics of an election year affects the allocation of government funds in
response to the Covid-19 crisis across states, congressional districts, and industries in the U.S., and
the corresponding consequences for employment and business activity.
Our paper lies in the intersection of two voluminous literatures. The first studies the effect of
government spending on economic outcomes during periods of economic stress (e.g., Clemens and
Miran 2010; Chodorow-Reich et al. 2012; Wilson 2012; and Fishback and Kachanovskaya 2015).
The second studies the link between politics and government spending (e.g., Ritt 1976; Ray 1980,
1981; Kiel and McKenzie 1983; Atlas et al. 1995; Levitt and Poterba 1999; Sapienza 2004; Dinc
2005; Hoover and Pecorino 2005; Faccio, Masulis, and McConnell 2006; Aghion et al. 2009;
Cohen, Coval, and Malloy 2011; Duchin and Sosyura 2012, Goldman, Rocholl, and So 2013;
Adelino and Dinc 2014; Tahoun 2014; Tahoun and van Lent 2019; Schoenherr 2019; Brogaard et
al. 2020). We add to these literatures by emphasizing the role of election-year politics and political
22
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 2
0-65
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
polarization in the government’s response to the historic Covid-19 crisis and its economic
consequences.
To investigate the role of electoral politics in the allocation of government funds, we introduce
novel measures of states’, districts’ and industries’ political importance in an election year. Our
main hypothesis is that electoral political considerations tilted the allocation of PPP funds by the
Trump administration towards firms in areas or industries that can have a significant impact on the
results of the 2020 elections. The first set of measures aims to identify battleground states,
congressional districts, and sectors. Prior research shows that presidential campaigns strategically
concentrate their resource allocation in battleground areas (e.g., Bartels 1985; Shaw 1999; James
and Lawson 1999; Shachar and Nale-buff 1999; Panagopoulos 2006; Shaw 2008; Akey, Dobridge,
Heimer, and Lewellen 2018). We extend this research by studying battleground allocation of
emergency government funds in an election year.
To identify battleground states, we collect detailed data on political ad expenditures by the
Trump campaign and by third parties, which are collectively higher in states with more competitive
elections. In particular, we collect data on political ad spending on Facebook, and measure the
proportion of the Trump campaign’s Facebook ad spending across states. We also collect data
published by the Federal Election Commission (FEC) on state spending by third parties, which are
not affiliated with any candidate, and measure the percent of third-party funding in opposition to
Donald Trump’s presidential campaign in each state. A combined higher proportion corresponds
to more competitive ad spending, indicating that the state is perceived as more important by the
Trump reelection campaign and by third-party political operatives.
To identify battleground congressional districts, we use the most recent Partisan Voting Index
(PVI) produced by the Cook Political Report. The PVI uses data from the last two presidential
23
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 2
0-65
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
elections to determine the voting performance of a congressional district relative to the national
average. Battleground districts are those with a PVI between D+10 and R+10. Lastly, we identify
battleground sectors based on the partisan industry classification of Gimpel, Lee and Parrott (2014,
henceforth GLP), which uses a decade of campaign contributions to congressional candidates by
corporations and trade associations.
The second set of measures aims to identify strategic political favoritism. According to this
view, the combination of identity politics and strident political polarization gives rise to a strategic
motive to allocate resources disproportionately to subgroups associated with the party’s base since
the outcome of elections is largely determined by the ability of politicians to mobilize base voters
rather than swing or opposition voters.2 These analyses extend existing research on political
favoritism in the allocation of non-emergency government funds in the U.S. outside election years
(e.g., Grossman 1994; Larcinese, Rizzo, and Testa 2006; Berry, Burden, and Howell 2010).
To measure strategic political favoritism at the state level, we use the most recent version of
the Cook Political National Report preceding the passage of the CARES Act (March 9, 2020). This
report categorizes states according to their likely voting outcome in the 2020 presidential election.
We classify a state as Republican if it is identified as “Likely Republican” or “Solidly Republican.”
At the congressional district level, we classify a district as Republican if the PVI is greater than
R+10. At the industry-level, we identify Republican sectors as those in the top tercile of
Republican leaning according to GLP.
In the first set of analyses, we investigate the determinants of the allocation of PPP loans across
states, districts, and sectors in the U.S. Since the allocation of PPP loans is an equilibrium outcome
of both supply and demand, the analyses consider the demand for PPP loans by controlling for
2 See, for example, Bernstein (2005) and Brown-Dean (2019) for an overview of identity politics and its rise in the United States.
24
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 2
0-65
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
state- or sector-level applications for PPP loans. The analyses are also adjusted for population size
or aggregate eligible payroll, depending on data availability, because the PPP’s primary focus is
on supporting employment through businesses’ payroll expenses.
At the state-level, we find that battleground states and Republican states receive more PPP
capital. Adjusted for a state’s aggregate eligible payroll expenses, an increase of one standard
deviation in battleground political ad spending corresponds to an increase of 2.9 percentage points
in the allocation of PPP loans, or an increase of 4.7% relative to sample mean. Furthermore,
Republican states receive 9.6 percentage points more PPP capital compared to other states, or
15.4% more relative to the sample mean. We find similar results at the congressional district and
sector levels. On a per capita basis, electorally important districts – battleground and Republican
districts – receive 20% and 12.7% more PPP loans, respectively, compared to other districts.
Similarly, scaled by total eligible payroll, electorally important sectors receive roughly 30% more
PPP loans relative to the sample mean.
These effects hold jointly, are highly statistically significant, and persist after controlling for
population size, the number of confirmed Covid-19 cases, unemployment claims at the onset of
the Covid-19 crisis, state-level gross domestic product (GDP) growth rates just before the onset of
the crisis, and the presence of banks with historical ties to flagship SBA loan programs. The
findings also hold for an aggregate index of electoral importance that combines the individual
measures. We also investigate the demand for PPP loans and show that it does not vary with
electoral importance, suggesting that credit demand is not driving the effects. Collectively, these
estimates suggest that electoral politics plays an important role in the provision of emergency
government funding during an election year, highlighting the strategic importance of both swing
and base voters.
25
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 2
0-65
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
We also consider the hypothesis that the effects are exacerbated by the loose monitoring and
continuously changing terms of the PPP.3 To test this hypothesis, we exploit the staggered
implementation of the PPP. We argue that the public outcry that followed the initial stages of the
PPP led to an increase in scrutiny and public attention to the PPP between the first and second
rounds of the program.4 Consequently, we expect the effect of electoral politics on credit provision
to weaken between the rounds. Consistent with this hypothesis, the estimates show that electoral
politics plays a weaker role in the second round of the PPP. The effect of electoral politics on the
allocation of second round PPP loans is economically small and mostly statistically insignificant.
Overall, this evidence is less consistent with the view that the measures of electoral importance
capture credit demand or underlying economic conditions correlated with the allocation of the
PPP, such as the economic exposure of states to the Covid-19 crisis, which a-priori should not
change between the two rounds of the PPP.
In the second set of analyses, we provide evidence on the real economic effects of the allocation
of PPP loans. First, we provide two-stage-least-squares (2SLS) estimates using data from the Small
Business Pulse Survey (SBPS).5 The survey is conducted by the U.S. Census Bureau and provides
high-frequency information on the impact of Covid-19 on small businesses and on the participation
of small businesses in government programs such as the PPP. In the first-stage regression, we
predict the allocation of PPP loans using the measures of electoral importance. In the second-stage
regressions, we investigate the effects of the predicted PPP allocation on the reported economic
impact of Covid-19 on small businesses.
3 See, for example: “House Passes Bill Loosening Rules on PPP Small-Business Loans” by Natalie Andrews and Amara Omeokwe, https://www.wsj.com/articles/community-lenders-to-get-10-billion-of-ppp-small-business-loans-11590678108. 4 See, for example, “Ruth’s Chris to Repay Loan Amid Outcry Over Rescue Program” by Peter Rudegeair, Heather Haddon, and Ruth Simon, https://www.wsj.com/articles/public-companies-have-to-repay-small-business-rescue-loans-11587670442 5 See: https://portal.census.gov/pulse/data/#about for a detailed description of this survey.
26
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 2
0-65
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
The 2SLS estimates suggest that the election-year allocation of PPP loans mitigates the
negative effects of Covid-19 on small business activity and employment. The estimated effects are
statistically significant and economically meaningful. An increase of 10% in predicted PPP
allocation corresponds to a decrease of 8.5% in the percentage of survey respondents who report
a negative effect of Covid-19 on their business and a decrease of 10% in the percentage of survey
respondents who temporarily close. Similarly, it corresponds to a decline 11.2% in reported
employment reductions. Overall, small businesses that received election-year PPP allocations are
considerably more likely to expect a quick return to normal operations.
Second, we provide estimates from difference-in-differences tests of business applications and
employment, where the first difference is between electorally important and all other states or
districts and the second difference is before versus after the onset of the first round of the PPP.
The estimates suggest that following the onset of the PPP, the decline in business applications was
attenuated by 2.79-9.51% in electorally important regions. Further, the increase in unemployment
was attenuated by 16.86% and the declines in aggregate employment and employment per capita
were attenuated by 5.3% and 1.26%, respectively. These effects hold after controlling for state,
week, or month fixed effects, as well as the interactions of the PPP time indicator with loan
demand, population size, GDP growth rate, and the presence of SBA banks. In contrast, we do not
find significant effects in placebo tests around the announcement of a national public health
emergency before the onset of the PPP.
Collectively, our findings suggest that election-year political considerations tilted the
allocation of emergency government funds in response to the Covid-19 crisis towards businesses
in electorally important states, districts, and industries. Our results add to the literature pioneered
by Stigler (1971) and Peltzman (1976) that studies how politics influences economic policy. These
allocational tilts have important real effects on business activity and employment, which could
impact the results of the 2020 elections.
27
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 2
0-65
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
2. The Paycheck Protection Program (PPP)
The Coronavirus Aid, Relief, and Economic Security (CARES) Act was passed by Congress with
overwhelming, bipartisan support and signed into law by President Trump on March 27th, 2020.
In total, the CARES Act designated over $2 trillion dollars to combat the adverse economic impact
of the Covid-19 pandemic, amounting to 10% of total U.S. gross domestic product (GDP), making
it the largest economic relief package in the history of the United States.
The Paycheck Protection Program (PPP) is a centerpiece $659 billion business loan program
established by section 1102 of the CARES Act, which authorized the Small Business
Administration (SBA) to distribute loans to support payroll and overhead expenses to eligible
small businesses through its nationwide network of lenders. Lenders who already participated in
the SBA’s flagship 7(a) program were automatically eligible to disburse PPP loans, while other
lenders had to obtain authorization from the SBA.
Each PPP loan is guaranteed by the SBA and loan applicants did not need to provide any
collateral or personal guarantees to apply or be approved for a PPP loan. Participating lenders
earned an upfront origination fee proportional to the amount of the loan: 5% for loans under $350k,
3% for loans between $350k and $2 million, and 1% for loans above $2 million.
The PPP focuses on small businesses, and, as such, eligibility for the PPP is based on the
existing statutory and regulatory definition of a “small business concern” under section 3 of the
Small Business Act, 15 U.S.C. 632. A business can qualify if it meets the SBA employee-based or
revenue-based small business size standard corresponding to its primary industry. Alternatively, a
business can qualify for the PPP if it meets the SBA’s “alternative size standard,” which requires
a maximum tangible net worth of $15 million and maximum average net income for the two full
fiscal years before the date of the application of $5 million.
28
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 2
0-65
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
The terms of PPP loans are highly attractive for the borrower. First, the principal of a PPP loan
can be either partially or fully forgiven based on the usage of the loan proceeds. Second, even if
not forgiven, PPP loans carry a low interest rate of one percent. Third, both the principal and
interest payments are deferred until the loan is forgiven or, if the borrower does not apply for loan
forgiveness, ten months after the end of the 24-week cover period.6 Consequently, millions of
businesses in the U.S immediately applied for PPP loans, which were accepted, approved, and
disbursed on a first-come first-served basis, leading to credit rationing and generating a setting
susceptible to political favoritism.7
The first round of the PPP commenced on April 3, 2020 amidst government-mandated
lockdowns in many states. Within 2 weeks, on April 16, 2020, the entire first round of $349 billion
was depleted and the SBA stopped accepting new applications from lenders.8 A bill to add $310
billion of funding was passed by Congress and signed into law by President Trump on April 24,
and the SBA began accepting new applications from lenders on April 27. The PPP was due to
expire at midnight on June 30 with funds remaining, but just hours before the expiration of the
program Congress authorized an extension through August 8. This date passed without a second
extension to the program, with the result that applying to the program is no longer possible. By the
end of the program, the SBA disbursed $525 billion of the $659 billion appropriated by Congress
to this program. These numbers indicate stark differences in the demand for loans between the two
rounds of the PPP: First-round PPP capital was quickly depleted, whereas second-round PPP
6 The SBA initially required that at least 75% of the loan be used for payroll, rent, mortgage interest, and utilities to be forgiven at the end of 8 weeks. On June 5, President Trump signed the PPP Flexibility Act, which reduced the proportion needed to be spent on payroll to 60% and extended the time period to use the funds from 8 to 24 weeks. 7 While the SBA did not release information about the number of PPP applications or application approval rates, it reported a total of 4.67 million loans disbursed by June 20, 2020. 8 See, for example, the article “Small business rescue loan program hits $349 billion limit and is now out of money,” by Thomas Franck and Kate Rogers, published on CNBC on April 16: https://www.cnbc.com/2020/04/16/small-business-rescue-loan-program-hits-349-billion-limit-and-is-now-out-of-money.html
29
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 2
0-65
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
capital exceeded aggregate demand. Hence, the differences between the two rounds provide a
natural setting to study the relation between political favoritism and credit rationing. 9
Moreover, we conjecture that in addition to lower demand, the second round of the PPP was
also accompanied by more stringent oversight, potentially reducing the scope for political
favoritism in loan allocation. In particular, the first round was followed by public outcry
surrounding the participation of large or public firms in the first round of the PPP.10 Moreover,
several lawsuits brought against J.P. Morgan Chase, Wells Fargo, Bank of America, and U.S. Bank
by a range of California small businesses further alleged that the banks unfairly prioritized their
large customers.11 In a press briefing on April 22, 2020, Treasury Secretary Mnuchin warned of
“severe consequences” for large businesses that received PPP funds.12 Following Mnuchin’s press
briefing, the SBA instituted a “safe harbor” for the return of PPP funds by large businesses, and
on April 28, the Treasury and SBA issued a joint statement that they would retroactively examine
all loans over $2 million to certify that program qualifications were met.13 We therefore
hypothesize that the apparent differences in both demand and oversight between the two rounds of
the PPP provide a natural backdrop against which to examine the impact of electoral politics on
government funding amid changing oversight credit rationing conditions.
9 See “Tracker: Paycheck Protection Program Loans,” by Thomas Wade: https://www.americanactionforum.org/research/tracker-paycheck-protection-program-loans/ 10 See, for example, the article “At Least 30 Public Companies Say They Will Keep PPP Loans,” by Inti Pacheco, published by the Wall Street Journal on May 19, 2020: https://www.wsj.com/articles/at-least-30-public-companies-say-they-will-keep-ppp-loans-11589891223 11 See, for example, the article “Chase and other banks shuffled Paycheck Protection Program small business applications, lawsuit says,” by Dalvin Brown, published in USA Today on April 20: https://www.usatoday.com/story/money/2020/04/20/small-businesses-sue-chase-bank-over-handling-stimulus/5163654002/. For further details on the lawsuits, pleas see: https://www.classaction.org/news/class-actions-say-wells-fargo-jpmorgan-chase-held-back-small-businesses-paycheck-protection-program-funds 12 See https://www.businessinsider.com/treasury-mnuchin-consequences-big-companies-taking-ppp-small-business-loans-2020-4 13 See https://factba.se/sba-loans for the list of public PPP borrowers, including those that subsequently returned the funds. The full PPP loan-level data can be found here: https://home.treasury.gov/policy-issues/cares-act/assistance-for-small-businesses/sba-paycheck-protection-program-loan-level-data.
30
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 2
0-65
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
3. Data and Variables
In this section, we describe our data sources and the construction of the variables used in the
analyses. We begin by describing the measures of PPP loan allocation at the congressional district,
state, and sector levels. We then describe our measures of electoral importance across districts,
states, and sectors. We conclude by describing measures of credit demand, credit supply, and local
economic conditions.
3.1. The Allocation of PPP Loans
We measure the allocation of PPP loans across congressional districts, states, and sectors, and
scale it by aggregate measures of eligible payroll when available because the primary goal of the
PPP was to support payroll expenses. These data come from the SBA, which provides detailed
data on PPP loans, and from the Statistics of U.S. Businesses (SUSB), which provides detailed
payroll data.
To study the allocation of PPP loans across states and sectors, we use aggregate state-level and
sector-level loan data released by the SBA for each week of the PPP. We use the SUSB payroll
data to estimate the total amount of payroll that is eligible for PPP funds within a particular state
or sector. In particular, we use the latest edition of the SUSB (2017) and calculate the total annual
payroll for all firms in NAICS sector 72 (Accommodation and Food Services) and for all firms
with 500 employees or fewer in all other sectors.14 Next, we aggregate these totals at the state or
sector level (adjusted to 2019 dollars) and divide them by 12 to calculate the aggregate monthly
payroll. Lastly, we multiply the monthly payrolls by 2.5 to approximate the procedure used by the
SBA to determine maximum PPP loan amounts, which aim to cover 2.5 months of payroll
expenses.
14 Firms in the Accommodation and Food Services were exempt from the 500-employee PPP eligibility cap.
31
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 2
0-65
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
PPP Funding Obs Mean p25 p50 p75 SDState
First-round PPP ($millions) 50 $6,803.89 $2,006.86 $4,465.70 $8,721.17 $6,852.59Second-round PPP ($millions) 50 $3,357.13 $522.36 $1,672.79 $3,790.20 $5,508.32Total PPP ($millions) 50 $10,161.02 $2,519.39 $6,359.24 $12,249.02 $12,013.29First-round PPP/Elig. Payroll 50 0.625 0.54 0.623 0.711 0.121Second-round PPP/Elig. payroll 50 0.222 0.154 0.202 0.275 0.081Total PPP/Elig. Payroll 50 0.848 0.814 0.85 0.882 0.066District
First-round PPP per capita 428 4.908 2.816 4.64 6.68 2.667Second-round PPP per capita 428 9.256 5.835 8.117 11.378 5.196Sector
First-round PPP/Elig. Payroll 18 0.547 0.438 0.529 0.665 0.184Second-round PPP/Elig. payroll 18 0.251 0.179 0.266 0.319 0.097
Political Measures Obs Mean p25 p50 p75 SDState
Trump Facebook ad share 50 0.02 0.007 0.017 0.025 0.019Third-party spend share 50 0.008 0 0 0.001 0.032Battleground state 50 0.018 0.006 0.013 0.02 0.021Republican state 50 0.42 0 0 1 0.499Electorally important state 50 0.433 0.333 0.333 0.667 0.263District
Republican district 428 0.287 0 0 1 0.453Battleground district 428 0.444 0 0 1 0.497Electorally important district 428 0.731 0 1 1 0.444Sector
Battleground sector 18 0.333 0 0 1 0.485Republican sector 18 0.278 0 0 1 0.461Electorally important sector 18 0.611 0 1 1 0.502
Local Economic Conditions Obs Mean p25 p50 p75 SDState
Eligible payroll ($millions) 50 $12,937.76 $3,210.72 $7,982.23 $16,135.03 $15,923.70Ln(population) 50 15.206 14.399 15.332 15.846 1.025Unem. per capita (04/04/2020) 50 0.044 0.032 0.04 0.055 0.017GDP growth 50 0.02 0.016 0.021 0.024 0.008% Small SBA lenders 50 0.106 0.041 0.075 0.177 0.081Ln(Covid-19 cases) (04/03/2020) 50 7.4 6.292 7.341 8.5 1.467District
Ln(population) 428 13.515 13.492 13.513 13.539 0.049Unem. Rate 428 0.038 0.031 0.036 0.042 0.009GDP growth 428 0.028 0.019 0.027 0.034 0.015% Small SBA lenders 428 0.07 0.019 0.042 0.101 0.074Ln(Covid-19 cases) (04/03/2020) 428 5.236 3.923 5.312 6.627 1.999
Survey Responses Obs Mean p25 p50 p75 SDState
% Applied to PPP (04/30/2020) 50 0.736 0.713 0.743 0.772 0.044Neg. effect on business 50 0.475 0.428 0.47 0.515 0.076Temp. closed business 50 0.381 0.327 0.37 0.433 0.085Return to normal <= 1 month 50 0.031 0 0.035 0.051 0.027Return to normal > 6 month 50 0.296 0.262 0.296 0.332 0.051Sector
% Applied to PPP (04/30/2020) 18 0.72 0.625 0.744 0.827 0.169
Real Effects Variables (per capita) Obs Mean p25 p50 p75 SDTotal business applications 850 0.21 0.15 0.186 0.232 0.103Corporation applications 850 0.024 0.012 0.016 0.024 0.022High propensity applications 850 0.071 0.051 0.063 0.078 0.034Continued unem. Claims 800 0.029 0.006 0.015 0.047 0.028Employment 2,764 0.026 0.008 0.019 0.042 0.021
Table 1: Summary Statistics
32
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 2
0-65
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
Figure 1: The Allocation of PPP Loans across Congressional Districts, States, and Sectors
Panel A: Congressional Districts
Panel B: States
Panel C: Sectors
33
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 2
0-65
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
This figure shows the allocation of first-round PPP loans across congressional districts, states, and sectors in the United States. Panel A reports the number of first-round loans per capita across congressional districts. Panel B reports aggregate loan amounts scaled by eligible payroll across states. Panel C reports aggregate loan amounts scaled by eligible payroll across sectors, defined based on 2-digit NAICS.
To study the allocation of PPP loans across congressional districts, we cannot use aggregate
loan data because they are only available at the state and sector levels. Instead, we use loan-level
data released by the SBA on 7/6/2020. The availability of these data, however, varies by the
amount of the loan. For loans over $150k, the data include the name, address, 6-digit NAICS
industry, business type (sole proprietorship, corporation, etc.), number of jobs retained, and a range
for the loan amount. The ranges for the loan amounts are as follows: $150-$350k, $350k-$1mil,
$1mil-$2mil, $2mil-$5mil, and $5mil-$10mil. For loans under $150k, the name of the business is
suppressed, and the loan amount is exact. Since loan amounts are imprecise and district-level
payroll data are largely unavailable, we measure the allocation of PPP loans across congressional
districts based on the number of loans scaled by the size of the population.
Table 1 shows that, on average, nearly 63% of state-level eligible payroll, and 55% of sector-
level eligible payroll, were covered by the first round of PPP loans. At the district level, the average
number of first-round PPP loans per 1,000 residents was 4.908, with a median of 4.64. The
allocation of PPP loans is also depicted in Figure 1. The heat maps in Figure 1 show substantial
variation in the allocation of PPP loans across districts (Panel A) and states (Panel B). In particular,
Panel B shows that states in the Midwest and South received more PPP funding relative to their
eligible payroll, while states on the coasts received relatively less. Panel C of Figure 1 shows the
nontrivial variation in the allocation of PPP across sectors.
3.2. Electoral Importance
In this sub-section we describe our measures of electoral importance. We measure electoral
importance via base (Republican) and swing (Battleground) voters across congressional districts,
34
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 2
0-65
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
states, and sectors. Each unit of analysis utilizes unique data sources to quantify the extent to which
districts, states, and sectors are important for the outcome of the 2020 presidential election.
Republican Districts, States, and Sectors
To measure the support of congressional districts for the Republican party, we utilize the Partisan
Voting Index (PVI) provided by the Cook Political Report. The PVI measures how each
congressional district’s presidential voting results compare to the national average based on the
previous two presidential elections. For example, a PVI value of R+10 indicates that the district
voted 10 points more Republican in the 2012 and 2016 elections, on average, than the national
average. We use the latest edition of the PVI as of 2017.15 We define an indicator variable,
Republican district, which equals one if the PVI is greater than R+10. Table 1 shows that roughly
29% of the congressional districts are Republican districts.
To measure the support for the Republican party across states, we use the most recent version
of the Cook Political Report preceding the passage of the CARES Act (March 9, 2020). This report
categorizes states according to their likely voting outcome in the 2020 presidential election. We
define an indicator variable, Republican state, which equals one if a state is identified as “Likely
Republican” or “Solidly Republican,” and zero otherwise. As shown in Table 1, 42% of the states
are Republican states based on the above definition.
We identify Republican 2-digit NAICS sectors using the partisan classification of Gimpel, Lee
and Parrott (henceforth GLP) (2014). GLP examine a decade of campaign contributions made by
corporations’ and trade associations’ political action committees to congressional candidates, and
pinpoint which industries have a measurable preference for a particular political party.
Importantly, their method identifies industries’ political preferences after controlling for other
15 See https://cookpolitical.com/introducing-2017-cook-political-report-partisan-voter-index for a detailed description of the PVI.
35
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 2
0-65
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
factors that likely drive campaign contributions, including parties’ majority control of Congress,
committee memberships, and the competitiveness of congressional seats. GLP then aggregate
industries to the 2-digit NAICS level, and compute the percent of the industries within each 2-digit
NAICS sector that favor a particular party (See GLP, Table 2).
We define an indicator variable, Republican sector, which equals one for sectors in the top
tercile on Republican leanings. We provide a list of Republican sectors in Appendix Table B.
Importantly, GLP do not report data on the Construction sector, which is a major participant in the
PPP. Therefore, we augment the GLP data with contributions data by sector from the Center for
Responsive Politics, and note that the Construction sector gave roughly 70% of its contributions
to Republican candidates in the 2018 election cycle. We therefore classify the Construction sector
as Republican, and note that our results are not sensitive to this inclusion. The estimates in Table
1 suggest that roughly 28% of the 2-digiti NAICS sectors are Republican sectors.
Battleground Districts, States, and Sectors
We identify battleground congressional districts using the PVI. In 2016, 23 districts with
Republican representatives voted for Hillary Clinton with margins ranging from 0.6 to 19.7. On
the other hand, 12 districts with Democrat representatives voted for Donald Trump with margins
ranging from 0.7 to 30.8. To accommodate the wide range of battleground districts, we define an
indicator variable, Battleground district, which equals one if the PVI is between D+10 and R+10.
This definition provides a sufficient range to capture the congressional districts that are most “up
for grabs” in the 2020 presidential elections. Table 1 shows that roughly 44% of congressional
districts are Battleground districts. The variation in electoral importance across congressional
districts is depicted in Panel A of Figure 2, which presents a heat map of the PVI across districts.
36
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 2
0-65
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
To identify battleground states, we calculate the proportion of both Facebook ad spending by
the Trump campaign and expenditures by third-party political operatives in support of and in
opposition to Donald Trump. This measure captures the level of competition in political ad
spending. Moreover, the perceived electoral importance of states, as captured by the revealed
preferences of political campaigns and operatives, is likely a potent instrument for electoral
importance since it drives allocative decisions. We define the variable Battleground state as the
share of Trump Facebook ad spending and third-party ad spending in each state.
We note that advertising spending in the 2020 presidential race has already reached
unprecedented levels. According to the Media Project at Wesleyan University, total spending for
the 2020 presidential elections had already eclipsed 2016 spending by Feb. 23, 2020. Further,
digital political advertising is becoming increasingly important. Industry experts project an
increase of 203% relative to 2016 for political digital marketing expenditures, versus an 82%
increase for traditional TV advertising. Facebook has become the preferred digital platform for all
politicians since it allows targeted advertising of individuals based on geographic location.16
Digital advertising is particularly important for the Trump campaign, which has devoted 70%
of its advertising resources to digital advertising, of which the majority went to Facebook.17 We
estimate the variable Trump Facebook ad share using data provided by the Facebook Transparency
Project as compiled by the Campaign 2020 Tracker.18 It is defined as the share of the Trump
campaign’s total Facebook ad spending that goes to a particular state from March 30, 2019 (when
the data is first available) to March 31, 2020. The estimates in Table 1 show that the average and
median Trump Facebook ad share are 2% and 1.7%, respectively. There is considerable variation
16 See: “Facebook accounts for 60% of all digital political advertising,” https://www.emarketer.com/newsroom/index.php/us-political-ad-spending-to-hit-record-high/ 17 See the Wesleyan Media Project: http://mediaproject.wesleyan.edu/releases-022620/ 18 CampaignTracker 2020 data: https://2020campaigntracker.com/
37
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 2
0-65
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
across states, however. The minimum share is 0.2%, the maximum share is 9.1%, and the standard
deviation across states equals 1.9%.
We collect data on third-party (independent) political expenditures from the Federal Elections
Commission (FEC). The Code of Federal Regulations (CFR) defines an independent expenditure
as “an expenditure by a person for a communication expressly advocating the election or defeat of
a clearly identified candidate that is not made in cooperation, consultation, or concert with, or at
the request or suggestion of, a candidate, a candidate’s authorized committee, or their agents, or a
political party committee or its agents.” The FEC requires independent expenditures to be reported
within 24-48 hours, and records the name of the spender, the location (state) of the spending, and
whether it is in support of, or in opposition to, a particular candidate.
We calculate third-party political expenditures, Third-party spend share, as the state share of
total independent expenditures supporting or opposing Donald Trump from January 2019 to March
2020. In the 2020 election cycle, the majority of third-party ad expenditures were in opposition to
Donald Trump (roughly $23 million compared to $16 million). This measure captures the relative
focus of third-party political operatives on a particular state. Table 1 shows that the median Third-
party spend share is 0, indicating that most states did not have third-party political expenditures
expressly supporting or opposing Donald Trump. The variation across states is nontrivial,
however, with a maximum share of 19.2% and a standard deviation of 3.2%. Overall, third-party
political operatives spent money in 19 states in the leadup to the PPP.
Lastly, we identify battleground 2-digit NAICS sectors using data from GLP (2014). We define
an indicator variable, Battleground sector, which equals one for sectors whose Republican
leanings are in the middle tercile and zero otherwise. A list of battleground sectors can be found
in Appendix Table B.
38
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 2
0-65
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
Figure 2: The Electoral Importance of Congressional Districts and States Sectors
Panel A: Congressional Districts
Panel B: States
This figure shows the electoral importance of congressional districts, states, and sectors for the 2020 presidential elections in the United States. Panel A reports the Partisan Voting Index across congressional districts. Panel B reports the Electorally important index, defined based on relative political ad spending and the Republican dummy, across states.
Electorally Important Index
We also construct a composite index of electoral importance at the district, state, and sector levels
by combining the above elements. We define districts and sectors as Electorally important if they
are either Republican or battleground districts/sectors. Similarly, we define states as Electorally
important if they are either battleground or Republican after transforming the continuous variable
Battleground state into an indicator variable that equals one if it is above the sample median, and
taking the average of the two variables. Panel B of Figure 2 presents a heat map of the variation in
the Electorally important index across states.
39
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 2
0-65
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
3.3. Local Supply and Demand of PPP Loans
To measure the local supply of PPP loans, we collect data on SBA 7(a) lenders, which, as noted
above, were immediately eligible to disburse PPP loans. In particular, we hand-match
comprehensive SBA 7(a) loan data as of Dec. 31, 2019 to bank branch locations from the FDIC
Summary of Deposits database, and compute the proportion of local (state or district) branches
operated by SBA banks. We conjecture that access to PPP loans was easier in areas with greater
presence of SBA lenders, especially in the first round. Furthermore, both anecdotal and academic
evidence suggest that bank size played a role in access to PPP loans (e.g., Granja et al., 2020; Liu
and Volker, 2020). Specifically, small, community banks were better able to navigate the labor-
intensive PPP application system and obtain funds for their clients. Hence, we proxy for the local
supply of PPP loans using the presence of small SBA banks (<$1 billion in assets). Table 1 shows
that the average share of small SBA banks is 10.6% across states and 7.0% across districts, with
large variation across both states and sectors (standard deviations = 8.1% and 7.4%, respectively).
To measure the demand for PPP loans, we utilize survey data provided by the Census Bureau.
These data come from the Small Business Pulse Survey (SBPS), which was initiated to track the
effects of the coronavirus and subsequent government interventions on small businesses. The
target population for the survey is all nonfarm, single-location businesses with less than 499
employees. Although the sample for the SBPS is not a random sample, weights are applied to
ensure that each weekly sample represents the full population of businesses. The SBPS conducts
weekly email surveys that began on 4/26. We focus on the first survey, which catalogued responses
40
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 2
0-65
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
as of 4/30. Our proxy for loan demand is the percent of state or sector respondents that reported
applying to the PPP since 3/13/2020.19
Table 1 shows that an average of 73.6% of state survey respondents applied to the PPP,
suggesting that a majority of small businesses in the United States applied for government aid. The
interquartile range for PPP demand across the states is relatively small. The 25th percentile equals
71.3%, and the 75th percentile equals 77.2%. This lack of variation in the demand across states
provides suggestive evidence that the variation in our political measures across states does not
simply proxy for state-level demand for PPP loans. Similarly, an average of 72% of industry
respondents applied to the PPP.
3.4. Economic Conditions
To control for the economic conditions within a particular congressional district or state, we
supplement our analyses with various local economic indicators. At the district level, we include
the weighted county-level unemployment rate as of 2019, GDP growth rates as of 2018, and the
natural log of the population size, where weights are determined by the proportion of a district’s
population that resides in a particular county.
At the state level, we include GDP growth rates as of the 4th quarter of 2019, unemployment
claims per capita as of the beginning of the first round of the PPP, and the natural log of the
population size. Data on GDP come from the Bureau of Economic Analysis (BEA). Data on
unemployment claims and rates come from the Bureau of Labor Statistics (BLS). Population data
come from the Census Bureau. All variables represent the latest available data before the beginning
of the first round of the PPP. Table 1 provides summary statistics for the above measures of local
economic conditions across states and congressional districts.
19 When analyzing the second round of the PPP, we use the most recent survey week (as of 6/25) to proxy for demand.
41
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 2
0-65
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
We also control for the local exposure to Covid-19 by including the natural log of the number
of Covid-19 cases as of the beginning of the PPP. These data are provided by USA Facts.
Lastly, we analyze the real effects of the allocation of PPP loans on business applications and
unemployment claims. Data on weekly business applications come from the Census Bureau. These
data report applications by businesses for an Employee Identification Number (EIN), and are
divided into three buckets: total applications, corporate applications, and high-propensity business
applications. Data on monthly sector employment by state come from the BLS. All variable
definitions can be found in Appendix A.
4. Results
We begin the empirical analyses by investigating the role of electoral importance in the allocation
of PPP loans across congressional districts, states, and sectors. We then examine the variation in
the demand for PPP loans and the differences between the first and second rounds of the PPP. We
conclude this section with an investigation of the real effects of the allocation of PPP loans on
business activity and employment using both survey evidence and observational data on local
economic conditions.
4.1. Electoral Importance and the Allocation of PPP Loans
To investigate the role of electoral importance in the allocation of PPP loans, Table 2 presents
estimates from cross-sectional regressions explaining the allocation of PPP loans across
congressional districts (columns 1-2), states (columns 3-4), and sectors (columns 5-6). These
regressions focus on the allocation of PPP loans in the first round of the program, when credit was
rationed and before the public outcry that led to more scrutiny and monitoring. Section 4.3 below
provides evidence on the second round of the PPP.
42
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 2
0-65
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
Since aggregate PPP loan volume data and accurate payroll information are unavailable or
noisy for congressional districts, we measure the allocation of PPP loans across districts by the
number of loans scaled by the size of the population (columns 1-2). These data are available for
states and sectors; hence, the dependent variables in columns 3-6 are the dollar volume of PPP
loans scaled by aggregate eligible payroll.
The main variables of interest in the regressions are Republican, Battleground, and Electorally
important, which measure the electoral importance of congressional districts, states, and sectors
based on their relative support for the incumbent administration (base voters), their electoral
competitiveness (swing voters), and the combination of the two, respectively. Depending on data
availability at the district, state, and sector levels, the regressions control for PPP loan demand
based on the SBPS (% Applied to PPP), local economic conditions (Unemployment, GDP growth),
exposure to the Covid-19 crisis (Ln(Covid 19 cases)), the availability of PPP lenders (% Small
SBA lenders), and population size (Ln(Population)).
We begin with an analysis of the allocation of PPP loans across congressional districts in
columns 1 and 2 of Table 2. Column 1 provides estimates for the individual measures Republican
and Battleground, whereas column 2 focuses on the composite index Electorally important. The
estimates in columns 1 and 2 suggest that electoral importance played an important role in the
allocation of PPP loans across districts. Republican districts and battleground districts received a
higher number of PPP loans, scaled by the size of the population. These effects are statistically
significant at the 1% and 5% levels, respectively, and hold after controlling for loan demand, local
economic conditions, and the availability of PPP lenders. The economic magnitude of the effects
is nontrivial. Relative to Democratic districts, Republican districts received 12.78% more loans
per capita in the first round of the PPP, and battleground districts received 20.01% more loans.
43
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 2
0-65
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
Overall, the estimates in column 2 show that electorally important districts received 14.51% more
loans per capita in the first round of the PPP.
This table examines the effect of electoral importance on the allocation of first-round PPP loans across congressional districts (columns 1-2), states (columns 3-4), and sectors (columns 5-6). The dependent variable in columns 1-2 is the number of PPP loans in a district scaled by its population size. The dependent variable in columns 3-4 and 5-6 is the aggregate amount of PPP loans in a state or sector, respectively, scaled by eligible payroll. All variable definitions are given in Appendix A. Heteroskedasticity-robust t-statistics are reported in parentheses. ***, **, and * denote significance at the 1%, 5%, and 10% levels, respectively.
We obtain similar results in columns 3 and 4, which study the allocation of PPP loans across
states. The estimates suggest that electorally important states received a higher dollar volume of
PPP loans relative to their aggregate levels of eligible payroll. In particular, the results indicate
Obs. Level Cong. District Cong. District State State Sector SectorColumn (1) (2) (3) (4) (5) (6)
Republican 0.982*** 0.096*** 0.16-2.725 -3.953 -1.496
Battleground 0.627** 1.394*** 0.169**-2.068 -4.634 -2.53
Electorally important 0.712** 0.150*** 0.165**-2.401 -4.736 -2.52
% Applied to PPP 3.346 3.734 0.993*** 1.021*** 0.568** 0.0565**-0.658 -0.735 -3.344 -3.569 -2.37 -2.44
Unemployment -81.321*** -81.480*** -0.016 -0.022**(-6.809) (-6.875) (-1.464) (-2.258)
Ln(Covid 19 cases) 0.001 -0.024 -0.040** -0.049***-0.022 (-0.402) (-2.427) (-3.167)
Ln(population) -2.98 -2.743 -0.661 -0.89(-1.121) (-1.028) (-0.878) (-1.505)
GDP growth -18.921** -18.721** 2.176** 2.317**(-2.507) (-2.443) -2.215 -2.226
% Small SBA lenders 13.339*** 13.504*** 0.710*** 0.693***-7.866 -7.85 -5.897 -5.529
Constant 44.747 41.416 0.460* 0.631*** 0.038 0.04-1.263 -1.163 -1.881 -2.768 -0.204 -0.223
Observations 428 428 50 50 18 18R-squared 0.305 0.302 0.805 0.793 0.431 0.431
Table 2: The Allocation of PPP Loans
44
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 2
0-65
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
that in the first round of the program, Republican states received 15.36% more funding per eligible
payroll than Democratic states, and battleground states received 4.68% more funding. Overall,
based on column 4, electorally important states received 7.82% more funding than electorally
unimportant states in the first round of the PPP. These findings are highly statistically significant
at the 1% level.
Finally, columns 5 and 6 of Table 2 provide the results for the allocation of PPP loans across
sectors. Despite the small number of observations (18 sectors), we find that electoral importance
played a statistically significant role in the allocation of PPP funds in the first round of the program.
The coefficient estimates show that battleground sectors received 30.9% more proportional
funding than Democratic sectors. The coefficient on Republican sectors is positive and of similar
magnitude to Battleground sectors, but is insignificant at conventional levels. The composite index
of political importance (column 6) remains highly economically and statistically significant.
Taken together, the results in this section suggest that political favoritism in an election year
operates through two distinct channels: base voters (Republican) and swing voters (Battleground).
Given that voters focus on recent economic outcomes (e.g., Achen and Bartels (2004)), the results
are consistent with the incumbent administration strategically tilting government funds towards
areas and industries that could play an important role in the 2020 presidential election.
These effects can operate via several distinct mechanisms. First, the incumbent administration
can pressure the SBA, formally or informally, to prioritize loan applications from electorally
important districts, states, and sectors. Second, the lending financial institutions can themselves
prioritize applications from electorally important borrowers to cater to the current administration.
Lastly, the program design itself may inherently favor electorally important borrowers.
45
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 2
0-65
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
4.2. The Demand for PPP Loans
A possible concern with the analyses is that the political importance of districts, states, and sectors
is correlated with the demand for PPP loans. Under this view, the role of electoral importance in
the allocation of PPP loans is driven by the demand for loans rather than by political favoritism.
We address this concern in several ways. First, the summary statistics in Table 1 show that there
is little variation in the demand for loans across states and sectors. Second, the regressions in Table
2 explicitly control for the demand for loans. In this section, however, we also seek to provide
direct evidence on the variation in the demand for loans across states and sectors (data on loan
applications are unavailable for congressional districts.)
In Table 3, we estimate predictive regressions explaining the demand for PPP loans in the first
round of the program across states (columns 1-2) and sectors (columns 3-4). The main takeaway
from Table 3 is that electoral importance in unrelated to the demand for PPP loans. Across all four
columns of Table 3, the estimates suggest that electoral importance is unrelated to the demand for
loans in the first round of the PPP. The coefficient estimates on Republican, Battleground, and
Electorally important are economically small, flip signs, and statistically insignificant at
conventional levels. Moreover, columns 1 and 2 show that the demand for PPP loans across states
is unrelated to any of the control variables, including local economic conditions and the exposure
to the Covid-19 crisis. As expected, the only exception is the size of population, which is positively
related to the aggregate demand for PPP loans.
Collectively, these results suggest that the effect of electoral importance on the allocation of
PPP loans in the first round of the program is not driven by variation in the demand for PPP across
electorally important states or sectors.
46
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 2
0-65
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
This table examines the effect of electoral importance on the demand for PPP loans across states (columns 1-2) and sectors (columns 3-4). The dependent variable is the percentage of Small Business Pulse Survey (SBPS) respondents in a state or sector that reported applying to the PPP by 4/30. All variable definitions are given in Appendix A. Heteroskedasticity-robust t-statistics are reported in parentheses. ***, **, and * denote significance at the 1%, 5%, and 10% levels, respectively.
4.3. The Second Round of the PPP
The analyses thus far focused on the first round of the PPP. In this section, we investigate the role
of electoral importance in the allocation of loans in the second round of the PPP. We conjecture
that the effects of electoral importance on loan allocation are driven by loose monitoring and credit
rationing. To test this conjecture, we exploit the differences between the two rounds of the PPP.
We argue that the public outcry that followed the initial stages of the PPP led to an increase in
Dependent VariableColumn (1) (2) (3) (4)
Republican 0.008 0.004-0.62 -0.038
Battleground 0.158 -0.053-0.867 (-0.520)
Electorally important 0.012 -0.027-0.76 (-0.317)
Ln(Covid-19 cases) 0.001 0.001-0.18 -0.108
Ln(population) 0.020* 0.020*-1.732 -1.904
Unemployment 0.42 0.41-0.958 -1.063
GDP growth -0.483 -0.485(-0.463) (-0.487)
% Small SBA lenders -0.049 -0.049(-0.656) (-0.681)
Constant 0.413** 0.421*** 0.736*** 0.736***-2.568 -3.06 -10.022 -10.35
Observations 50 50 18 18R-squared 0.303 0.3 0.025 0.007
State Demand Sector DemandTable 3: The Demand for PPP Loans
47
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 2
0-65
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
scrutiny and public attention to the PPP in its second round. This claim is supported by numerous
articles and actions taken by policy makers (see section 2 above). Furthermore, the supply of PPP
loans exceeded the demand for loans in the second round, suggesting that credit was not rationed.
Consequently, we expect the effect of electoral politics on credit provision to weaken between the
two rounds.
To test this conjecture, Table 4 repeats the analyses of the allocation of PPP loans (Table 2)
replacing the dependent variables with the allocation of loans in the second round of the PPP. We
also replace the measures of PPP loan demand with analogous measures using SBPS data as of
6/25/2020.
Consistent with our hypothesis, the results in Table 4 suggest that electoral importance did not
play a significant role in the allocation of loans in the second round of the PPPP across
congressional districts, states, and sectors. In particular, the coefficient estimates on the different
measures of electoral importance flip signs across specifications and are statistically insignificant
in the majority of cases (8 out of the 10 cases). When significant (2 out of 10 cases), they have a
negative sign.
Combined with the results on the allocation of loans in the first round of the PPP, these results
have two important implications. First, they suggest that omitted variables correlated with the
design of the PPP, which likely remained constant through both rounds of the PPP, cannot explain
the effects of electoral importance on the allocation of loans. Second, they indicate that lax
monitoring and credit rationing serve as key mechanisms and motivations in political favoritism.
The loosening of credit conditions and the increase in monitoring and public scrutiny reduced the
motivation and scope, respectively, for political favoritism in the allocation of loans in the second
round of the PPP.
48
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 2
0-65
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
This table examines the effect of electoral importance on the allocation of second-round PPP loans across congressional districts (columns 1-2), states (columns 3-4), and sectors (columns 5-6). The dependent variable in columns 1-2 is the number of PPP loans in a district scaled by its population size. The dependent variable in columns 3-4 and 5-6 is the aggregate amount of PPP loans in a state or sector, respectively, scaled by eligible payroll. All variable definitions are given in Appendix A. Heteroskedasticity-robust t-statistics are reported in parentheses. ***, **, and * denote significance at the 1%, 5%, and 10% levels, respectively. 4.4 Real Economic Effects
The evidence thus far suggests that the electoral importance of congressional districts, states, and
sectors played a role in the allocation of PPP loans. A natural question that arises is whether these
allocations had real economic consequences. In this section, we seek to provide this evidence by
Obs. Level Cong. District Cong. District State State Sector SectorColumn (1) (2) (3) (4) (5) (6)
Republican 0.399 -0.032 -0.062-0.534 (-1.473) (-1.481)
Battleground 0.712 -1.078*** 0.059-1.053 (-5.441) -1.382
Electorally important 0.637 -0.064** 0.003-0.965 (-2.094) -0.076
% Applied to PPP -6.149 -6.491 -0.003* -0.003* 0.004*** 0.003**(-0.764) (-0.811) (-1.700) (-1.786) -3.875 -2.354
Unemployment -0.963*** -0.961*** 0.003 0.003(-4.254) (-4.237) -0.221 -0.248
Ln(Covid 19 cases) 0.904*** 0.927*** 0.038** 0.039**-5.445 -6.073 -2.45 -2.259
Ln(population) 1.83 1.622 0.311 0.329-0.461 -0.41 -0.805 -0.971
GDP growth 0.355** 0.353** 0.261 0.082-2.166 -2.159 -0.271 -0.088
% Small SBA lenders -2.325 -2.471 -0.433*** -0.429***(-0.840) (-0.882) (-4.930) (-4.643)
Constant -13.223 -10.288 -0.124 -0.123 -0.005 0.024(-0.248) (-0.194) (-0.602) (-0.571) (-0.057) -0.217
Observations 428 428 50 50 18 18R-squared 0.189 0.188 0.666 0.643 0.535 0.293
Table 4: Second-Round Allocation of PPP Loans
49
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 2
0-65
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
utilizing survey evidence on small business activity as well as observational economic data on
business applications and employment.
4.4.1 Survey Evidence on Small Business Activity
We begin the investigation of real economic effects with evidence from responses to the Small
Business Pulse Survey. To capture the impact of electoral importance on the allocation of PPP
loans and consequently on real economic outcomes, we employ a two-stage-least-squares
approach. The first-stage regression estimates the effect of electoral importance on the allocation
of PPP loans. The second-stage regressions use the predicted allocation of PPP loans from the
first-stage to explain the variation in survey responses. The analyses focus on the first round of the
PPP, where the evidence shows that electoral importance played a role in the allocation of loans.
Furthermore, they focus on the variation in the allocation of PPP loans and survey responses across
states because the survey does not provide responses across congressional districts.
The analyses focus on survey responses to the following questions:
1) Overall, how has this business been affected by the COVID-19 pandemic?
2) In the last week, did this business temporarily close any of its locations for at least one
day?
3) In the last week, did this business have a change in the number of paid employees?
4) In your opinion, how much time do you think will pass before this business returns to its
usual level of operations?
We construct the outcome variables as the percent of survey responses to each of the above
questions in each state. For example, Neg. effect on business is the percent of survey respondents
in a state that answered “Large negative effect” in response to question 1. Temp. business closure
50
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 2
0-65
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
is the percent of respondents answering “Yes” to question 2. Appendix A provides the detailed
definition of each variable.
Table 5 reports these results. Column 1 provides estimates from the first-stage regression,
which show that electoral importance played a role in the allocation of PPP loans across states in
the first round of the program. This result is evident from the positive and statistically significant
coefficient on the composite index of electoral importance (first-stage F-statistic of 22.43).
Columns 2-6 report the second stage estimates of the regressions of surveyed small business
activity on predicted PPP funding in the first round. The evidence is consistent across all the
survey-based variables. Small businesses in states that received higher allocation of PPP loans, as
predicted by their electoral importance, were less likely to report a negative effect on their business
(column 2), less likely to temporarily close their business (column 3), less likely to reduce
employment (column 4), more likely to expect a return to normal in less than a month (column 5),
and less likely to expect a return to normal in more than 6 months (column 6). Qualitatively, these
results suggest that the allocation of PPP funds to electorally important states attenuated the
negative effects of the Covid-19 crisis on small business activity.
The economic magnitudes of these effects are meaningful. A 10-percentage point increase in
the predicted allocation of PPP loans decreases the percentage of survey respondents who report a
negative effect of Covid-19 on their business by 8.5%, the percentage of respondents who
temporarily closed their business by 10%, and the percentage of respondents reporting a decrease
in employment by 11.2%. Further, a 10-percentage point increase in predicted PPP allocation
increases the percent of respondents who expect a return to normal business operations in less than
1 month by 78.7% and decreases the percent of respondents who expect a return to normal business
operations in more than 6 months by 12.96%. Taken together, these results provide suggestive
51
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 2
0-65
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
evidence that the politically motivated allocation of PPP funds in the first round of the program to
electorally important states had important real economic effects for small businesses.
This table provides estimates from two-stage-least-squares regressions of the effect of electoral importance on the allocation of PPP loans and subsequently small business activity. The first-stage regression (column 1) predicts the allocation of first-round PPP loans across states using electoral importance. The second-stage regressions (columns 2-6) explain small business activity using the predicted values from the first-stage regressions. The dependent variables in the second-stage regressions are based on responses to the Small Business Pulse Survey. All variable definitions are given in Appendix A. Heteroskedasticity-robust t-statistics are reported in parentheses. ***, **, and * denote significance at the 1%, 5%, and 10% levels, respectively. 4.4.2 Difference-in-Difference Evidence on Business Applications and Employment
The survey-based analysis in Table 5 provides evidence from a single cross-section of states. In
the next set of analyses, we provide difference-in-differences estimates from panel regressions that
1st Stage
Dependent variable
First-round PPP
loans/Eligible payroll
Neg. effect on business
Temp. business closure
Reduce employm
ent
Return to normal <= 1
month
Return to normal > 6 months
Column (1) (2) (3) (4) (5) (6)
Electorally important 0.150***-4.736
-0.405*** -0.380** -0.283** 0.244** -0.374***(-2.987) (-2.337) (-2.214) -2.314 (-2.807)
% Applied to PPP 1.021*** 0.045 -0.019 -0.041 -0.366** 0.232-3.569 -0.164 (-0.069) (-0.202) (-2.171) -1.076
Ln(Covid-19 cases) -0.022** 0.020* 0.018 0.002 0.012** -0.011(-2.258) -1.9 -1.363 -0.324 -2.154 (-1.088)
Ln(population) -0.049*** -0.012 -0.021 0.002 0.014* 0.003(-3.167) (-0.966) (-1.257) -0.144 -1.648 -0.238
Unemployment -0.89 1.402** 2.197*** 1.790*** 0.245 1.027*(-1.505) -2.509 -3.688 -3.534 -0.771 -1.951
GDP growth 2.317** -0.965 -1.018 -0.65 0.568 0.974-2.226 (-1.227) (-1.196) (-1.122) -1.191 -0.965
% Small SBA lenders 0.693*** 0.038 -0.053 -0.065 -0.089 0.260*-5.529 -0.249 (-0.306) (-0.442) (-0.850) -1.707
Constant 0.631*** 0.684*** 0.746*** 0.359*** -0.166 0.297-2.768 -3.65 -3.638 -2.69 (-1.577) -1.619
Observations 50 50 50 50 50 50R-squared 0.793 0.612 0.582 0.612 0.098 0.241
2nd Stage
Table 5: Second-Round Allocation of PPP Loans
52
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 2
0-65
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
include state, sector, week, and month fixed effects, which alleviate concerns about unobservable
economic indicators and time trends that might confound the analyses.
We begin by analyzing the effect of PPP funding on the number of weekly business
applications per capita. If the allocation of PPP loans in the first round of the program matters for
economic recovery, we would expect that electorally important states experience higher business
applications following the onset of the PPP compared to less electorally important states. To test
this prediction, we construct a state-week panel from 1/4/2020 to 4/25/2020, and estimate the
following regression:
𝑌𝑠,𝑡 = 𝛽1𝐸𝑙𝑒𝑐𝑡𝑜𝑟𝑎𝑙𝑙𝑦 𝑖𝑚𝑝𝑜𝑟𝑡𝑎𝑛𝑡 ∗ 𝑅𝑜𝑢𝑛𝑑 1 + 𝛽2𝑋𝑠,𝑡 ∗ 𝑅𝑜𝑢𝑛𝑑 1 + 𝛾𝑠 + 𝛼𝑡 + 𝑒𝑠,𝑡
Where 𝑌𝑠,𝑡 is one of three measures of business applications per capita for state s in week t, Round
1 is a dummy variable equal to 1 after the beginning of the first round of the PPP (4/4/2020),
𝐸𝑙𝑒𝑐𝑡𝑜𝑟𝑎𝑙𝑙𝑦 𝑖𝑚𝑝𝑜𝑟𝑡𝑎𝑛𝑡𝑠,𝑡 is the composite index of electoral importance, and 𝑋𝑠,𝑡 contains all
the control variables used in our cross-sectional analyses. This specification allows us to control
for permanent differences between treatment (electorally important) states and control states, along
with aggregate time trends at the granular weekly level. Importantly, we allow all the explanatory
variables, including states’ electoral importance, loan demand, economic conditions, and exposure
to the Covid-19 crisis, to have a differential impact before and after the onset of the first round of
the PPP. This approach ensures that the observed impact of electoral importance during the first
round of the PPP is not due to a differential response of electorally important states along other
observable dimensions.
53
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 2
0-65
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
This table examines the effect of electoral importance on weekly new business applications and employment across states. Total business applications are the number of businesses applying for an employer identification number (EIN) scaled by population. Corporation business applications are the number of corporations applying for an EIN scaled by population. High propensity business applications are applications for an EIN that have a high likelihood of turning into businesses with a payroll. Cont. unem claims are weekly continued unemployment claims per capita. Ln(emp.) is log monthly employment by state and sector. Emp. per capita is monthly employment per capita by state and sector. All the regressions include state and week or month fixed effects. Columns 5-6 include sector fixed effects. Round 1 is a dummy variable equal to 1 from 4/4/2020 to 4/25/2020. All variable definitions are given in Appendix A. Standard errors are clustered at the state level. T-statistics are reported in parentheses. ***, **, and * denote significance at the 1%, 5%, and 10% levels, respectively.
We present these results in Table 6. The key explanatory variable is the interaction term
Electorally important * Round 1, which captures the differences across electorally important and
unimportant states following the onset of the crisis. The estimates show that following the onset
of the first round of the PPP, electorally important states experienced increased business
applications compared to unimportant states. These results are evident from the positive coefficient
on the interaction term Electorally important * Round 1. The results hold across the different
definitions of business applications and are economically nontrivial. Following the onset of the
Dependent variableTotal business
applications
Corporation business
applications
High propensity business
applications
Cont. unem claims
Ln(emp.)Emp. per capita
Column (1) (2) (3) (4) (5) (6)Electorally important * Round 1 0.018** 0.007** 0.007* -0.015** 0.052** 0.001**
-2.331 -2.423 -2.003 (-2.295) -0.022 -0.001% Applied to PPP * Round 1 0.039 0.021 0.001 0.02 -0.203 -0.001
-0.289 -0.924 -0.012 -0.343 -0.199 -0.004Ln(Covid-19 cases) -0.003** 0.000 -0.001 0.000 -0.005** -0.000***
(-2.566) (-0.717) (-0.786) -0.863 -0.002 0.000Ln(population) * Round 1 -0.002 -0.004** -0.002 0.001 0.002 0.000
(-0.290) (-2.079) (-0.813) -0.519 -0.011 0.000GDP growth * Round 1 -0.336 0.109 -0.074 -0.234 0.234 0.004
(-0.694) -0.934 (-0.411) (-0.890) -1.038 -0.022% Small SBA lenders * Round 1 0.062 0.021* 0.025 0.014 0.120 0.002
-1.435 -1.781 -1.507 -0.503 -0.098 -0.003
Observations 600 600 600 600 2,744 2,764R-squared 0.969 0.939 0.943 0.875 0.931 0.856State FE Yes Yes Yes Yes Yes YesSector FE Yes YesWeek FE Yes Yes Yes YesMonth FE Yes Yes
Table 6: Difference-in-Difference Evidence on Business Applications and Employment
54
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 2
0-65
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
first round of the PPP, A one standard deviation increase in the Electorally important index
increases total business applications per capita by 2.79%, corporate applications by 9.51%, and
high-propensity business applications by 3.21%.
Columns 4-6 of Table 6 provides estimates from a similar difference-in-differences analysis of
weekly continued unemployment claims and monthly employment rates by state and sector.
Column 4 shows that continued unemployment claims per capita rose less in electorally important
states following the onset of the first round of the PPP. This result is captured by the negative
coefficient on the interaction term Electorally important * Round 1 in column 4. Columns 5 and 6
show that state-by-sector declines in employment were attenuated by electoral importance
following the onset of the first round of the PPP. These results are captured by the positive
coefficients on the interaction terms Electorally important * Round 1 in columns 5 and 6.
Furthermore, the magnitudes of the effects are meaningful. A one standard deviation increase
in the Electorally important index attenuates the rise in continued unemployment claims per capita
by 16.86%, and the fall in log employment and employment per capita by 5.3% and 1.26%,
respectively, following the onset of the first round of the PPP. The estimates are also statically
significant at the 10% level or higher.
Collectively, the results suggest that the strategic allocation of emergency government funds
in an election year helped mitigate the deleterious effects of the Covid-19 crisis on employment,
and helped to spur economic recovery by promoting new business applications. Given voters’
tendency to focus on recent economic performance, these positive economic effects could impact
the results of the 2020 elections.
In Table 7, we address the remaining concern that the positive effects of states’ electoral
importance on business applications and employment are driven by contemporaneous factors that
55
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 2
0-65
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
are unrelated to the allocation of PPP loans. For example, electorally important states may have
responded better to the Covid-19 emergency. To address this concern, we exploit the granular
nature of the weekly business applications and continued unemployment claims to conduct placebo
tests around dates that coincide with the Covid-19 crisis and are unrelated to the onset of the PPP.
This table provides estimates from placebo difference-in-differences regressions that replace the allocation of first-round PPP loans (Round 1 in Table 6) with the declaration of a national public health state of emergency on 1/31/2020. Public health is an indicator variable that equals 1 in the 4 weeks following the declaration and 0 in the 4 weeks prior to the declaration. All the regressions include state and week fixed effects. All variable definitions are given in Appendix A. Standard errors are clustered at the state level. T-statistics are reported in parentheses. ***, **, and * denote significance at the 1%, 5%, and 10% levels, respectively.
In particular, we examine the relative response of electorally important states versus electorally
unimportant states around the declaration of a national public health emergency on January 31,
2020. If electorally important and unimportant states vary in their exposure or response to the
Covid-19 crisis, we should also observe differences in business applications and unemployment
Dependent variableTotal business
applications
Corporation business
applications
High propensity business
applications
Cont. unem claims
Column (1) (2) (3) (4)Electorally important * Public health 0.00 -0.001 -0.001 0
(-0.068) (-0.844) (-0.213) (-0.130)% Applied to PPP * Public health -0.121* -0.018 -0.03 -0.001
(-1.846) (-1.209) (-1.068) (-0.841)Ln(weekly Covid-19 cases) -0.001 0.00 0.00 0.00
(-0.730) -0.544 -0.004 -0.608Ln(population) * Public health 0.007 0.003** 0.004 0.00
-1.323 -2.052 -1.675 (-0.175)GDP growth * Public health -0.132 -0.027 -0.027 -0.005
(-0.525) (-0.451) (-0.226) (-0.635)% Small SBA lenders * Public health -0.066* -0.009 -0.027 0.00
(-1.990) (-1.445) (-1.611) -0.692
Observations 600 600 600 600R-squared 0.969 0.939 0.943 0.875State FE Yes Yes Yes YesWeek FE Yes Yes Yes Yes
Table 7: Placebo Tests
56
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 2
0-65
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
claims across electorally important and unimportant states around this date. We focus our analysis
on the 9 weeks surrounding the declaration of the public health emergency, to avoid an overlap
with the initiation of the PPP.
The results are reported in Table 7. The estimates show that around the declaration of a national
public health emergency, business applications and unemployment claims in electorally important
states were indistinguishable from those in electorally unimportant states. Together, these findings
mitigate concerns that unobservable state characteristics correlated with the allocation of PPP
loans to electorally important states are driving the difference-in-differences effect of the PPP on
business applications and unemployment claims.
5. Concluding Remarks
This paper investigates the impact of election-year politics on the allocation of emergency
government funds through the flagship Paycheck Protection Program (PPP), which disbursed
forgivable loans to small businesses in the United States in response to the Covid-19 crisis, and its
real economic consequences. We construct novel measures of electoral importance that aim to
capture strategic capital allocation to swing and base voters. These measures use data from
Facebook ad spending, independent political expenditures, the Cook Political Report, and
campaign contributions across states, congressional districts, and sectors.
We provide two main results. First, businesses in electorally important states, districts, and
sectors receive more government funds following the onset of the Covid-19 crisis, controlling for
funding demand and both health and economic conditions. Second, the tilt in government funding
weakens the adverse effects of Covid-19 on employment, business applications, and small business
57
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 2
0-65
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
activity. These estimates are corroborated by both small business survey data and aggregate
economic data released by the Bureau of Labor Statistics.
Collectively, these estimates provide novel evidence on the allocative distortions and real
effects of electoral politics. These findings have important implications for the design and
governance of government investment programs, suggesting that regulators and policy makers
should pay particular attention to the implementation of such programs during election years.
58
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 2
0-65
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
References
Achen, Christopher H., and Larry M. Bartels. “Blind retrospection: Electoral responses to drought, flu, and shark attacks.” (2004).
Adelino, Manuel, and I. Serdar Dinc. “Corporate distress and lobbying: Evidence from the Stimulus Act.” Journal of Financial Economics 114, no. 2 (2014): 256-272.
Aghion, Philippe, Leah Boustan, Caroline Hoxby, and Jerome Vandenbussche. “The causal impact of education on economic growth: evidence from US.” Brookings papers on economic
activity 1 (2009): 1-73. Akey, Pat, Christine Dobridge, Rawley Heimer, and Stefan Lewellen. “Pushing boundaries:
Political redistricting and consumer credit.” Available at SSRN 3031604 (2018). Alesina, Alberto, John Londregan, and Howard Rosenthal. “A model of the political economy of
the United States.” American Political Science Review 87, no. 1 (1993): 12-33. Atlas, Cary M., Thomas W. Gilligan, Robert J. Hendershott, and Mark A. Zupan. “Slicing the
federal government net spending pie: Who wins, who loses, and why.” The American
Economic Review 85, no. 3 (1995): 624-629. Bartels, Larry M. “Expectations and preferences in presidential nominating campaigns.” The
American Political Science Review (1985): 804-815. Bernstein, Mary. “Identity politics.” Annu. Rev. Sociol. 31 (2005): 47-74. Berry, Christopher R., Barry C. Burden, and William G. Howell. “The president and the
distribution of federal spending.” American Political Science Review (2010): 783-799. Brogaard, Jonathan, Matthew Denes, and Ran Duchin. “Political influence and the renegotiation
of government contracts.” Review of Financial Studies (forthcoming). Brown-Dean, Khalilah L. Identity Politics in the United States. John Wiley & Sons, 2019. Chodorow-Reich, Gabriel, Laura Feiveson, Zachary Liscow, and William Gui Woolston. “Does
state fiscal relief during recessions increase employment? Evidence from the American Recovery and Reinvestment Act.” American Economic Journal: Economic Policy 4, no. 3 (2012): 118-45.
Clemens, Jeffrey, and Stephen Miran. “The effects of state budget cuts on employment and income.” Working paper. (2010).
Cohen, Lauren, Joshua Coval, and Christopher Malloy. “Do powerful politicians cause corporate downsizing?.” Journal of Political Economy 119, no. 6 (2011): 1015-1060.
Dinç, I. Serdar. “Politicians and banks: Political influences on government-owned banks in emerging markets.” Journal of Financial Economics 77, no. 2 (2005): 453-479.
Duchin, Ran, and Denis Sosyura. “The politics of government investment.” Journal of Financial
Economics 106, no. 1 (2012): 24-48. Faccio, Mara, Ronald W. Masulis, and John J. McConnell. “Political connections and corporate
bailouts.” The Journal of Finance 61, no. 6 (2006): 2597-2635.
59
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 2
0-65
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
Fair, Ray C. “The effect of economic events on votes for president.” The Review of Economics
and Statistics (1978): 159-173. Fishback, Price, and Valentina Kachanovskaya. “The multiplier for federal spending in the States
during the Great Depression.” The Journal of Economic History 75, no. 1 (2015): 125-162. Gimpel, James G., Frances E. Lee, and Michael Parrott. “Business interests and the party
coalitions: industry sector contributions to US congressional campaigns.” American Politics
Research 42, no. 6 (2014): 1034-1076. Goldman, Eitan, Jörg Rocholl, and Jongil So. “Politically connected boards of directors and the
allocation of procurement contracts.” Review of Finance 17, no. 5 (2013): 1617-1648. Granja, João, Christos Makridis, Constantine Yannelis, and Eric Zwick. Did the Paycheck
Protection Program Hit the Target?. No. w27095. National Bureau of Economic Research, 2020.
Grossman, Philip J. “A political theory of intergovernmental grants.” Public Choice 78, no. 3-4 (1994): 295-303.
Hoover, Gary A., and Paul Pecorino. “The political determinants of federal expenditure at the state level.” Public Choice 123, no. 1-2 (2005): 95-113.
James, Scott C., and Brian L. Lawson. “The political economy of voting rights enforcement in America's Gilded Age: Electoral College competition, partisan commitment, and the federal election law.” American Political Science Review 93, no. 1 (1999): 115-131.
Kiel, Lisa J., and Richard B. McKenzie. “The impact of tenure on the flow of federal benefits to SMSA's.” Public Choice 41, no. 2 (1983): 285-293.
Kiewiet, D. Roderick. Macroeconomics and micropolitics: The electoral effects of economic
issues. University of Chicago Press, 1983. Kramer, Gerald H. “Short-term fluctuations in US voting behavior, 1896–1964.” American
Political Science Review 65, no. 1 (1971): 131-143. Larcinese, Valentino, Leonzio Rizzo, and Cecilia Testa. “Allocating the US federal budget to the
states: The impact of the president.” The Journal of Politics 68, no. 2 (2006): 447-456. Levitt, Steven D., and James M. Poterba. “Congressional distributive politics and state economic
performance.” Public Choice 99, no. 1-2 (1999): 185-216. Liu, Haoyang, and Desi Volker. Where Have the Paycheck Protection Loans Gone So Far?. No.
20200506. Federal Reserve Bank of New York, 2020. Panagopoulos, Costas. “Vested interests: interest group resource allocation in presidential
campaigns.” Journal of Political Marketing 5, no. 1-2 (2006): 59-78. Peltzman, Sam. “Toward a more general theory of regulation.” The Journal of Law and
Economics 19, no. 2 (1976): 211-240. Ray, Bruce A. “Congressional promotion of district interests: does power on the Hill really make
a difference?.” Political Benefits (1980): 1-36.
60
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 2
0-65
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
Ray, Bruce A. “Military committee membership in the house of representatives and the allocation of defense department outlays.” Western Political Quarterly 34, no. 2 (1981): 222-234.
Ritt, Leonard G. “Committee Position, Seniority, and Distribution of Government Expenditures.” Public Policy 24, no. 4 (1976): 463-489.
Sapienza, Paola. “The effects of government ownership on bank lending.” Journal of Financial
Economics 72, no. 2 (2004): 357-384. Schoenherr, David. “Political connections and allocative distortions.” The Journal of Finance 74,
no. 2 (2019): 543-586. Shachar, Ron, and Barry Nalebuff. “Follow the leader: Theory and evidence on political
participation.” American Economic Review 89, no. 3 (1999): 525-547. Shaw, Daron R. “The methods behind the madness: Presidential electoral college strategies, 1988-
1996.” The Journal of Politics 61, no. 4 (1999): 893-913. Shaw, Daron R. “The race to 270: The electoral college and the campaign strategies of 2000 and
2004.” University of Chicago Press, 2008. Stigler, George J. “The theory of economic regulation.” The Bell Journal of Economics and
Management Science (1971): 3-21. Tahoun, Ahmed. “The role of stock ownership by US members of Congress on the market for
political favors.” Journal of Financial Economics 111, no. 1 (2014): 86-110. Tahoun, Ahmed, and Laurence Van Lent. “The personal wealth interests of politicians and
government intervention in the economy.” Review of Finance 23, no. 1 (2019): 37-74. Tufte, Edward R. Political control of the economy. Princeton University Press, 1978. Wilson, Daniel J. “Fiscal spending jobs multipliers: Evidence from the 2009 American Recovery
and Reinvestment Act.” American Economic Journal: Economic Policy 4, no. 3 (2012): 251-82.
61
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 2
0-65
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
Appendix A
Variable Definitions
Variable Definition Source First-round PPP loans/Elig. payroll Total $ PPP funds allocated to a given state or
sector from 4/3-4/14, scaled by payroll of firms with less than 500 employees. Payroll
for all firms in NAICS sector 72 (Accommodation and Food Services) are also
included.
SBA, SUSB
Second-round PPP loans/Elig. payroll
Total $ PPP funds allocated to a given state or sector from 4/3-6/30 minus total $ PPP funds allocated from 4/3-4/14, scaled by payroll of firms with less than 500 employees. Payroll
for all firms in NAICS sector 72 (Accommodation and Food Services) are also
included.
SBA, SUSB
First-round PPP per capita Total # PPP funds allocated to a given congressional district from 4/3-4/14, scaled by
district population
SBA, Census
Second-round PPP per capita Total # PPP funds allocated to a given congressional district from 4/3-6/30 minus total # PPP funds allocated from 4/3-4/14,
scaled by district population
SBA, Census
Republican state Likely or Solidly Republican state Cook Political Report (3/9/2020) Third-party spend share State share of third-party political spending in
opposition to and in support of Donald Trump (01/01/2019-03/31/2020)
FEC
Trump Facebook ad spending State share of Trump political ad spending on Facebook from 03/30/2019-04/04/2020
Facebook, Campaign Tracker 2020
Battleground state State share of total Trump Facebook ad
spending and Third-party spend share Facebook, FEC, Campaign Tracker
2020
Electorally important state Average of Republican state and a dummy variable for above-median Battleground state
Cook Political Report, Facebook/Campaign Tracker 2020,
FEC
Republican district Congressional districts that have a Partisan Voting Index of greater than R+10
Cook Political Report
62
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 2
0-65
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
Battleground district Congressional districts that have a Partisan Voting Index of between D+10 and R+10
Cook Political Report
Electorally important district Union of Republican district and Battleground district
Cook Political Report
Republican sector 2-Digit NAICS sectors in the top tercile of Republican leaning
GLP 2014
Battleground sector 2-Digit NAICS sectors in the middle tercile of Republican leaning
GLP 2014
Electorally important sector Union of Republican sector and Battleground
sector GLP 2014
% Applied to PPP % of Survey respondents that applied for PPP loan (as of 4/30 for first-round PPP or as of
6/27 for second-round PPP)
Small Business Pulse Survey
Ln(population) Natural log of population Census Ln(Covid-19 cases) The natural log of Covid-19 cases as of 4/03
or 4/25. Measured at the state and congressional district level.
USA Facts
Unemployment The sum of state continued unemployment claims and initial unemployment claims as of
4/04 or 4/25 for states, and population-weighted county unemployment rate from
2019 for districts
BLS
GDP growth State GDP 2019 Q4 growth. Defined as the population-weighted county GDP growth
from 2018 for congressional districts
BEA
% Small SBA lenders Proportion of branches of banks in a state or district under $1 billion in assets that
participated in the SBA 7(a) program from 2015-2019
SBA, Summary of Deposits
Total business applications Weekly Applications for an EIN Census Corporate business applications High-propensity business applications from a
corporation or personal service corporation, based on the legal form of organization stated
in the IRS Form SS-4
Census
63
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 2
0-65
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
High-propensity business applications
Business applications that have a high propensity of turning into businesses with a
payroll. High-propensity applications include applications: (a) from a corporate entity, (b)
that indicate they are hiring employees, purchasing a business or changing
organizational type, (c) that provide a first wages-paid date (planned wages); or (d) that
have a NAICS industry code in manufacturing (31-33), retail stores (44), health care (62), or
restaurants/food service (72)
Census
Continued unemployment claims Number of weekly state continued unemployment claims
DOL
Employment Total employment by state-sector-month. Measured in logs or per capita
BLS
Neg. effect on business The percent of survey respondents who reported a "Large negative effect" or
"Moderate negative effect" of Covid-19 on their business based on the 4/26 survey
SBPS
Temporary business closure The percent of survey respondents who reported temporary closing at least one
business location in the last week based on the 4/26 survey
SBPS
Reduced employment The percent of survey respondents who reported reducing employment in the last
week based on the 4/26 survey
SBPS
Return to normal <= 1 month The percent of survey respondents who predict a return to normal levels of operation in less than 1 month based on the 4/26 survey
SBPS
Return to normal > 6 Months The percent of survey respondents who predict a return to normal levels of operation
in more than 6 months based on the 4/26 survey
SBPS
64
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 2
0-65
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
Appendix B
A.2. List of Republican and Battleground Sectors
This list shows the NAICS sectors designated as having Republican preference in the top tercile (Republican sectors) and middle tercile (Battleground sectors) according to historical congressional campaign contributions. Partisan preference by sector comes from Gimpel, Lee, and Parrott (2014) and the Center for Responsive Politics.
NAICS Sector NAICS Description
Republican Sectors
11 Agriculture, Forestry, Fishing and Hunting 21 Mining, Quarrying, and Oil and Gas Extraction 23 Construction
31-33 Manufacturing 72 Accommodations and Food Service 52 Finance and Insurance
Battleground Sectors
44-45 Retail Trade 48-49 Transportation and Warehousing
81 Other Services 22 Utilities 51 Information 53 Real Estate and Rental and Leasing
65
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 2
0-65
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
Covid Economics Issue 62, 18 December 2020
Copyright: Javier G. Gómez-Pineda
Growth forecasts and the Covid-19 recession they convey: End-2020 update1
Javier G. Gómez-Pineda2
Date submitted: 14 December 2020; Date accepted: 15 December 2020
The paper updates the results in Gómez-Pineda (2020) about the depth, length and shape of the covid-19 recession using information up to the December-2020 forecast vintage. The method is a decomposition of output between potential output and the output gap, the former explained by supply shocks and the later, by demand shocks. We find that, compared to the July-2020 forecast vintage, in the December-2020 forecast vintage the median depth of the recession improved 1.1 percentage points in advanced economies while deteriorated 2.3 percentage points in emerging and developing economies. This change in the outlook may be explained by the increase in the prevalence of the disease in the second half of 2020 in emerging and developing economies as well as by the more limited reach of monetary and fiscal policies in emerging and developing economies. The recession is still V-shaped with partial recovery in advanced economies and L-shaped in emerging and developing economies. The results point to the relevance and urgency of policies to support emerging and developing economies.
1 This paper is an update of “Growth forecast and the Covid-19 recession they convey” published in issue 40 of Covid Economics in July 2020. The author thanks David Felipe López for excellent research assistance. The findings, recommendations and interpretations expressed in this paper are those of the author and do not necessarily reflect he view of the Banco de la República or its Board of Directors.
2 Senior Economist (Investigador Principal) at Banco de la República and part time professor (profesor de cátedra) at Universidad del Rosario.
66
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 6
6-73
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
1. Introduction
In the December forecast vintage, the projected depth of the recession improved in advanced
economies and deteriorated in emerging and developing economies. In advanced economies the
median depth of the recession improved about 1.1 percentage points to –7.9 percent, from –9.1
percent in July.1 In emerging and developing economies the median depth of the recession
deteriorated about 2.3 percentage points to –8.8 percent, from nearly –6.4 percent in July.2 In the
near future the risks in advanced economies are to the upside, as massive vaccinations can turn
mandatory and voluntary distance measures less necessary. In contrast, in emerging market and
developing economies the risks are to the downside, as the number of cases and deaths raise and
delays in vaccinations prolong the need for mandatory and voluntary distance measures.
At the time the July forecast vintage was made, the multiyear projections could be considered
relatively reliable for the year 2020 but not as reliable for the medium term. As suggested by the
forecast record during the Global Financial Recession, to give an example, medium term
projections, particularly those for emerging and developing economies, had major revisions (see
Gómez-Pineda, 2020b).
The covid-19 recession is explained by a combination of demand and supply shocks. In
advanced economies supply shocks appear to be better explained as output level shocks while in
emerging and developing economies they appear to be better explained as output growth shocks
(Gómez-Pineda, 2020a).
In this paper we update the results in Gómez-Pineda (2020a) using the information available
at end-2020. Section 2 presents a brief verbal description of the model, section 3 mentions the
main features of the latest available data, section 4 presents the updated results and section 5
concludes.
1 Figures may not add up because of rounding. 2 These numbers can change with the measure of central tendency; that is, whether the median or the weighted average is used; with the definition of distance, whether logarithmic or percent distance is used; and also with the sample of economies in the study. The reported figures correspond to the median, the logarithmic deviation and are based on a sample of economies described in the data section.
67
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 6
6-73
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
2. The model
The model consists of a breakdown of output between potential output and the output gap. Potential
output follows supply shocks while the output gap follows demand shocks. Supply shocks can be
output level, output growth shocks or a combination of both. Output level shocks affect the output
level permanently while output growth shocks affect the potential-output growth rate about the
long-term, potential-output growth rate temporarily. The long-term, potential-output growth rate
is obtained as the forecasted output growth at the end of the forecasting horizon. This forward-
looking measure of the long-term, potential-output growth rate might be more relevant for the
recession that is being projected compared with a rate obtained from past data.
Detrended output is obtained subtracting trend potential output; the later is obtained as the
potential output that obtains in the absence of supply shocks. Trend potential output is normalized
at 0 in the base year 2019.
3. The data
The sample of economies is the one available to us in our Focus Economics service; that is, 65
economies, 29 advanced and 36 emerging and developing.3 In 2019, the economies in the sample
accounted for 83.5 percent of world output evaluated at PPP exchange rates, 38.7 for advanced
economies and 44.8 for emerging and developing economies. Excluding China, the emerging and
developing economies account for 25.6 of world output. A more detailed description of the
database is available in Gómez-Pineda (2020a).
Most of this paper compares the December 2020 with the July 2020 forecast vintages. In the
Focus Economics source, the December forecast vintage includes data from November 8 through
November 22. In turn, the July forecast vintage includes data from June 14 through June 28.4
3 The countries in the sample are, in the group of advanced economies, Australia, Austria, Belgium, Canada, Cyprus, Estonia, Finland, France, Germany, Greece, Hong Kong SAR, Ireland, Italy, Japan, Korea, Latvia, Lithuania, Luxembourg, Netherlands, New Zealand, Portugal, Singapore, Slovak Republic, Slovenia, Spain, Switzerland, Taiwan Province of China, United Kingdom and United States; and in the group of emerging and developing economies, Argentina, Bangladesh, Belize, Bolivia, Brazil, Brunei Darussalam, Cambodia, Chile, China, Colombia, Costa Rica, Dominican Republic, Ecuador, El Salvador, Guatemala , Haiti, Honduras, India, Indonesia, Jamaica, Lao P.D.R., Malaysia, Mexico, Mongolia, Nicaragua, Pakistan, Panama, Paraguay, Peru, Philippines, Russia, Sri Lanka, Thailand, Trinidad and Tobago, Uruguay and Vietnam. 4 Because the information for Latin America and the Caribbean is the latest available, for the economies in this region we use the forecast vintage of the previous month.
68
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 6
6-73
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
Importantly, most of the December forecast vintage did not include any of the news about the
effectiveness of vaccines that came out on November 17. Future forecast vintages might show how
the news about the effectiveness, approval and distribution of vaccines affect growth forecasts.
4. The information in the end-2020 forecast vintage
In the December forecast vintage, the depth of the recession in 2020 improved in advanced
economies and deteriorated in emerging market and developing economies (Figure 1). The new
projections can be considered relatively reliable for 2020, as they are now based on observed data
for up to the second and third quarters of the year. In contrast, the projection of the depth of the
recession in the medium term, as said above, may not be as reliable. Medium term projections can
change importantly, for example, as mentioned above, current medium term projections are subject
to massive upside and downside risks.
Figure 1. A deeper recession is projected for emerging and developing economies Median output and median potential output in the Focus Economics December and July 2020 forecast vintages
Source: authors calculations based on data from Focus Economics.
In order to gauge the shape of the recession, a detrended method may be used. Some analysts
prefer to compare current projections with pre covid-19 projections. I prefer to use the rate of
growth currently projected for the end of the forecasting horizon because that is the new steady
state that will have to be reached during the recovery from the ongoing recession. Using this
method, the projected recession, with information up to end-2020, remains V-shaped with partial
recovery in advanced economies and L-shaped in emerging market and developing economies
(Figure 2). The shape of the recession is broadly maintained across quartiles (Figure 2). In addition,
80
85
90
95
100
105
110
115
120
2016 2018 2020 2022 2024
A. Advanced economies
Median output (December vintage)
Median output (July vintage)
Median potential output 90
95
100
105
110
115
120
2016 2018 2020 2022 2024
B. Emerging and developing economies
Median output (december vintage)
Median output (July vintage)
Median potential output
69
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 6
6-73
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
the extent of the interquartile range, with the more standard economies, is larger in emerging
market and developing economies (Figure 2).
Figure 2. The projected depth and shape of the covid-19 recession Quartile distribution of detrended output in the Focus Economics December 2020 forecast vintage
Source: authors estimations based on data from Focus Economics.
The new information in the December-2020 forecast vintage is about the depth of the
recession. Indeed, back in July 2020 growth forecasts tended to be negatively correlated with per
capita income (Figure 3, Panel A). In contrast, in December 2020, growth forecasts tend to be
positively correlated with per capital income (Figure 3, Panel B). As a result, the prospects for
advanced economies have improved while those for emerging and developing economies have
deteriorated (Panel C).
Figure 3. Growth forecast deteriorated in emerging market and developing economies PPP-weighted, per capita GDP vs. growth forecasts in the July and December 2020 forecast vintages
Source: authors calculations based on data from Focus Economics and John Hopkins University.
The deeper projected recession in emerging market and developing economies may be
explained by the increase in the prevalence of the disease in the second half of 2020, as gauged by
the number of deaths and cases (Figure 4). It can also be explained by the much more limited reach
-18
-14
-10
-6
-2
2
6
10
2016 2018 2020 2022 2024
B. Emerging and developing economies
First and fourth quartiles
Second and third quartiles
Median detrended output-18
-14
-10
-6
-2
2
6
10
2016 2018 2020 2022 2024
A. Advanced economies
First and fourth quartiles
Second and third quartiles
Median detrended output
-20-18-16-14-12-10
-8-6-4-20
0 20,000 40,000 60,000 80,000 100,000 120,000
2020
out
put g
row
th, j
une
fore
cast
vin
tage
Per capita GDP
A. July forecast vintage
-20
-18
-16
-14
-12
-10
-8
-6
-4
-2
00 20,000 40,000 60,000 80,000 100,000 120,000
2020
out
put g
row
th, D
ecem
ber f
orec
ast v
inta
ge
Per capita GDP
B. December forecast vintage
-12
-10
-8
-6
-4
-2
0
2
4
6
0 20,000 40,000 60,000 80,000 100,000 120,000
C. July - December forecast revision
70
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 6
6-73
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
of fiscal and monetary policies, see for instance Alberola et al. (2020), Cavallino and De Fiore
(2020) and Deb et al. (2020).
Figure 4. Number of cases and deaths Number of deaths and cases in the countries in the sample
Source: authors calculations based on data from John Hopkins University.
In our methodology, we work backwards from the growth forecasts to the shocks that may
be explaining the recession. Here we maintain the assumption we used by mid-2020 about the
distribution of the drop in output between supply and demand shocks. We maintain that the
distribution of the shocks between demand and supply is half and half only because of the large
uncertainty underlying this assumption. Some studies suggest demand shocks are larger while
others suggest the opposite, see Baqaee and Farhi (2020), Balleer et al. (2020) and see also the
overview in Macaulay and Surico (2020). Nonetheless, regardless the assumed distribution of
shocks, the result in Gómez-Pineda (2020a) about the likely type of supply shock in advanced and
emerging and developing economies is maintained in the December forecast vintage. The result is
that in advanced economies supply shocks are better characterized as output level shocks while in
emerging and developing economies they are better characterized as output growth shocks.
Using the updated information, the size of the output level shock in advanced economies in
2020 is smaller while the size of the output growth shock in 2020 in emerging and developing
economies is larger (Figure 5). The same applies demand shocks.
-
100,000
200,000
300,000
400,000
500,000
600,000
700,000
A. Number of deaths
Advanced economies Emerging and developing economies
-
5,000,000
10,000,000
15,000,000
20,000,000
25,000,000
30,000,000
B. Number of cases
Advanced economies Emerging and developing economies
71
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 6
6-73
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
Figure 5. The projected output level and output growth shocks in the covid-19 recession Median output-level and growth shocks in the Focus Economics July 2020 and December 2020 forecast vintages
Source: authors estimations based on data from Focus Economics.
5. Conclusion
The new information contained in the end-2020 growth forecasts is that in advanced economies
the projected recession is less deep while in emerging and developing economies the recession is
deeper. The deeper recession in emerging and developing economies may be explained by a larger
prevalence of the disease during the second half of 2020 as well as by the much more limited reach
of fiscal and monetary policies.
The risks to the outlook are to the upside in advanced economies, as massive vaccinations
are expected to contain the prevalence of the disease. In turn, in emerging market and developing
economies the risks to the outlook are to the downside, as delays in vaccinations may not help
contain the raise in the prevalence of the disease.
The updated information points to the relevance and urgency of policies to support emerging
and developing economies.
References
Alberola, Enrique, Yavuz Arslan, Gong Cheng and Richhild Moessner, “The Fiscal Response to the
COVID-19 Crisis in Advanced and Emerging Market Economies,” Bank for International
Settlements, BIS Bulletin N° 23, 2020.
Balleer, Almut, Sebastian Link, Manuel Menkhoff and Peter Zorn, (2020), “Demand or supply? Price
adjustment during the Covid-19 pandemic,” Covid Economics 31, 23 June.
-5-4-3-2-1012
2016 2018 2020 2022 2024
B. Emerging and developing economies: output growth shocks
July vintage
December vintage
-5-4-3-2-1012
2016 2018 2020 2022 2024
A. Advanced economies:output level shocks
July vintage
December vintage
72
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 6
6-73
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
Baqaee, David and Emmanuel Farhi, “Supply and Demand in Disaggregated Keynesian Economies with
an Application to the COVID-19 Crisis,” NBER Working Paper, No. 27152, 2020.
Cavallino, Paolo and Fiorella De Fiore, “Central Banks’ Response to COVID-19 in Advanced
Economies,” Bank for International Settlements, BIS Bulletin N° 21, 2020.
Deb, Pragyan, Davide Furceri, Jonathan Ostry and Nour Twak, “The Economic Effects of COVID-19
Containment Measures,” Covid Economics, CEPR Press, Issue 24, 2020.
Gómez-Pineda, Javier, “Growth forecasts and the covid-19 recession they convey.” Covid Economics,
CEPR Press, Issue 40, July, 2020a.
Gómez-Pineda, Javier, “The depth, length and shape of the covid-19 recession conveyed in mid-2020
growth forecasts,” Borradores de Economía, no. 1123, 2020b.
Macaulay, Alistair and Paolo Surico, “Is the Covid-19 recession caused by supply or demand factors?”
Economics Observatory, 2020.
73
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 6
6-73
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
Covid Economics Issue 62, 18 December 2020
Copyright: Claudius Gros and Daniel Gros
The economics of stop-and-go epidemic control1
Claudius Gros2 and Daniel Gros3
Date submitted: 11 December 2020; Date accepted: 13 December 2020
We analyse ‘stop-and-go’ containment policies which produce infection cycles as periods of tight lock-downs are followed by periods of falling infection rates, which then lead to a relaxation of containment measures, allowing cases to increase again until another lock-down is imposed. The policies followed by several European countries seem to fit this pattern. We show that ’stop-and-go’ should lead to lower medical costs than keeping infections at the midpoint between the highs and lows produced by ’stop-and-go’. Increasing the upper and reducing the lower limits of a stop-and-go policy by the same amount would lower the average medical load. But, increasing the upper and lowering the lower limit while keeping the geometric average constant would have the opposite impact. We also show that with economic costs proportional to containment, any path that brings infections back to the original level (technically a closed cycle) has the same overall economic cost.
1 This research has received funding from the European Union’s Horizon 2020 research and innovation program, PERISCOPE: Pan-European Response to the Impacts of Covid-19 and future Pandemics and Epidemics, under grant agreement No. 101016233, H2020-SC1-PHE-CORONAVIRUS-2020-2-RTD. We thank Charles Wyplosz and an anonymous reviewer for valuable comments. This research was motivated by a discussion of the second author with Jakob von Weizsaecker.
2 Professor, Institute for Theoretical Physics, Goethe-University Frankfurt.3 Distinguished Fellow, CEPS (Centre for European Policy Studies).
74
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 74
-89
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
1 Introduction
As governments grapple with the second wave in Europe, they are usually takinga much more gradual and graduated approach than during the initial phase ofthe pandemic. At that time the number of seriously ill increased so rapidly thatit overwhelmed health systems and in particular hospital capacities in severalcountries. The second wave, which started with the onset of the flu season in theautumn of 2020, has so far led to a somewhat moderated challenge for healthsystems.
After the first peak, the urgency to ‘flatten the curve’ [1] has subsided toa certain extent. However, governments still feel the need to take measuresto slow down the spread of the virus when the medical load is high. One keystrategic issue facing the authorities is whether they should try to preserve astatus quo, or whether to alternate lock-downs with periods of easing.
In Italy and England, the central governments have instituted a tiered sys-tem with different levels of social distancing restrictions. In regions with a higherincidence of COVID-19, the restrictions are tighter. Within such a ’traffic light’system a region (city or other subdivision) can graduate to a lower level of re-strictions if its epidemiological parameters improve, and, vice versa, restrictionswill be tightened if cases increase again. These countries have thus adopted defacto a ’stop and go’ policy at the regional level.
A change into a higher or lower category will of course become more fre-quent the closer the parameters defining the various tiers are. One key issuefor this ’regional traffic light’ approach is thus how wide apart these parametersshould be set. We investigate this issue keeping in mind that social distancingmeasures have an economic cost, which increases with their severity. The choiceof parameters should be informed by their economic cost, relative to the healthbenefits in terms of lower infections, hospitalizations and deaths [2].
We do not consider a general optimal control problem. Our aim is limitedto comparing policies that make intuitive sense and that describe the choicesof different European countries. At the national level one can observe thatGermany’s curve is relatively flat, compared to that of France, Belgium or Spain.See Figure 1.
There is one simple economic argument that would favour the ‘stop and go’strategy of alternating harsh restrictions with wide easing. The economic costof closing restaurants, closing schools or imposing restrictions on movements isthe same whether the current rate of infections is high or low. This implies thatone should use harsh restrictions when the case count is high because one wouldthen achieve the largest fall in cases (in absolute numbers).
The argument against the ‘stop and go’ strategy is that the cost of the harshrestrictions to achieve a quick fall in infection is likely to be convex. A smallproportional reduction in infections (or rather the reproduction number) canbe achieved by measures which have little impact on the economy (e.g. maskwearing, etc.). Achieving a more rapid deceleration in the diffusion of the virusrequires substantially stronger restrictions of the type mentioned above.
One could of course argue that stop and go policies are inferior to the ‘East
75
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 74
-89
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
Figure 1: Examples of second-wave control strategies. For the controlof the second Covid-19 wave in Europe, most countries imposed comparativelystrict lock-downs. Illustrative examples are, as shown, France, Belgium andSpain. As an exception, Germany, starting around the beginning of November2020 opted for a ‘semi-lockdown’, which resulted in near constant infections.Graphic generated using the Goethe Interactive Covid-19 Analyser [6].
Asian’ option of eradicating the virus [3], which then allows a total reopening ofthe economy. But this option has been abandoned in Europe as the draconianmeasures, including border closures that would be required, have apparentlybeen widely judged as unacceptable. Note, however, that it is currently yet tobe settled whether additional factors, like evolutionary adaptions, contributeto differences in the path the spread of the SARS-CoV-2 pathogen took inEuropean and East-Asian populations [4].
The remainder of our contribution is organised as follows: We start by brieflyreviewing the standard SIR model to which we add a relationship which de-scribes the economic cost of reducing the spread of the virus. This frameworkis then used to examine the economic control costs of two alternative policies:keeping the medical load constant [5] versus a stop and go policy. We thencompute the medical load implied by these two policies over a given time pathand compare the resulting relative economic costs against the benefits in termsof a lower overall number of infected. Finally, we consider the implications of atime varying native reproduction factor, for example an increase due to colderweather, leading to more indoor interactions.
Throughout, our purpose is not to describe and solve a general optimalcontrol problem, but to compare the economic cost of different concrete policyoptions. Figure 2 illustrates schematically the ’stop-and-go’ epidemic controlwhich we model below.
76
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 74
-89
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
2 Modelling framework
We start with a short presentation of the standard SIR model where we denotewith S = S(t) the fraction of susceptible (non-affected) people, with I = I(t)the fraction of the population that is currently ill (active cases, which are alsoinfectious), and with R = R(t) the fraction of recovered. Normalization de-mands S + I + R = 1 at all times. We write the continuous-time SIR modelas
τ S = −gSI, τ I =(gS − 1
)I, τR = I , (1)
which makes clear that τ is a characteristic time scale. Normalization is con-served, as S + I + R = 0. Infection and recovery rates are g/τ and 1/τ . Thenumber of infected grows as long as I > 0, namely when gS > 1. Herd immunityis consequently attained when the fraction of yet unaffected people dropped toS = 1/g. The total number of past and present infected is X = 1− S = I +R.
From (1) one sees that g/τ governs the transition between two compart-ments, from susceptible to infected. This transition rate is constant within thebasic version of the SIR model. There are two venues to relax this condition:
• Non-linear reproduction rates. The basic reproduction factor g maydepend functionally on the actual number of infected I [7, 8], or on thetotal number X, [2]. This happens when societies react on an epidemicoutbreak.
• Time-dependent reproduction rates. From the viewpoint of thepathogen, certain changes in the transmission rate g = g(t) are external,e.g. because hosts decide more often to quarantine.
Here we focus on time-dependent g = g(t), mostly as induced by stop and gopolitics, as illustrated in Figure 2. We assume a stop and go cycle which repeatsafter T = Tup + Tdown:
g(t) =
{gup > 1 t ∈ [0, Tup]
gdown < 1 t ∈ [Tup, T ]. (2)
The time spans during which epidemics expands/contracts are respectively Tupand Tdown. The policy is cyclic if I(0) = I(T ), viz when the starting case numberis reached again.
2.1 Low incidence approximation
We assume that infection counts are substantially lower than the population,viz that I � 1. For example, even in a highly affected country like Italy,the total number of daily infections has rarely exceeded thirty thousand [9],which corresponds to less than 0.0005 of total population. The total number ofcases has reached 1.5 million, which is equivalent to S ≈ 0.975. The number
77
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 74
-89
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
0 1 2 3 4 5 6
time
0
2
4
6
8
dai
ly c
ases
I
tightening
relaxing
tightening
g(down) g(up)arithmetic
geometric
Ilarge
Ilow
Figure 2: Stop-and-go epidemic control. Containment policies that alter-nate periodically between Ilow and Ilarge. When relaxing, the number of dailycases I = I(t) expands at a rate g(up) ≡ gup > 1. Policies are tightened againwhen case number are too high. Daily case numbers then contract with a rateg(down) ≡ gdown < 1. In the example shown, up- and down times are Tup = 1and Tdown = 2. For a presumed time scale of one month the period Tup +Tdown
of the control cycle would be here three months. Also indicated are the arith-metic and the geometric means of Ilow and Ilarge, as used in Sect. 4.1. Note thatcontant control is recovered in the limit Ilow → Ilarge.
of susceptibles remains then close to one, essentially constant and we can setS → 1. Hence we need to deal only with
τ I =(g − 1
)I, I(t) = I(t0) e(t−t0)(g−1)/τ . (3)
In the quasi-stationary state we have simple exponential growth/decay. Forstop-and-go control the reproduction factor is piece-wise constant, which allowsus to evaluate explicitly the time evolution of case numbers, and with this theassociated economic costs.
3 Economic costs of disease control
The economic costs of imposing social distancing on a wider population, closingrestaurants or retail trade, are at the core of policy discussions. The key issuehere is how these costs vary with the social distancing measures (so-called Non-Pharmacological Interventions, NPI) imposed. They increase in severity frommask requirements, abstaining from travel or restaurant meals to more invasiveinterventions like closure of schools, lock-downs or curfews. Limiting social in-teraction necessarily reduces economic activity. This suggests that the economiccost of the social distancing, measured as the proportional loss of GDP, should
78
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 74
-89
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
increase with the reduction in the transmission rate described by g,
E = ceg0 − gg0
, (4)
where g0 is the native reproduction factor. Here we assumed that social distanc-ing costs are proportional to the percentage-wise reduction of the reproductionfactor, viz to (g0 − g)/g0.
Our basic assumption, that social distancing costs are proportional to the re-duction in the reproduction parameter, differs from the assumptions underlyingmatching models, as used in [10] or [11], which typically arrive at a quadraticrelationship between the economic cost and the reduction in contagion, whereas[12] postulates simply a convex cost curve.
A concrete example can illustrate the key mechanism behind the matchingframework, which usually assumes that the ’lock-down’ takes the form of theconfinement of a proportion of the population, and that contagion is possibleonly outside. If one half of the population has to stay at home, only the otherhalf can go out and get potentially infected. But the half which is not confinedwill find only one half of their potential partners outside, resulting in one fourthof the number of matches. Contagion should thus be reduced by a factor of fourwhen confining one half of the population.
In the matching framework, the economic cost is assumed to be propor-tional to the percentage of the population which is confined - not the numberof matches. This framework thus separates social activity (matches, meetingsof people) from economic activity, assuming that contagion is fostered onlythrough social activity. Another implicit assumption behind this view is thatthose who are not confined will not have longer meetings with the ones theystill find outside; or that those who are free to move accept to have only halfof matches, and do not decide to meet somebody else if their preferred matchis not available. These implicit assumptions are crucial. For example, conta-gion would only be halved if those not confined would meet twice as long withthe remaining matches they find. Some authors, e.g. [11] acknowledge theseconsiderations by allowing for different economies of scale in matching.
Our view of lock-down or rather social distancing is that governments man-date the closure of some part of the economy, in reality mainly the services sector(restaurants, bars, shops, etc.). This is different from a strict confinement of apart of the population. A restaurant which is closed (or limited in its openinghours) results in less value added created and diminishes at the same time thepotential for contagion. But the restaurant owners and their workers are notconfined, they can meet others. There is thus no quadratic effect in terms ofcontacts. We thus start from the assumption that economic activity involvesoccasions for contagion, implying that the loss of economic activity should bedirectly proportional to the reduction in occasions for contagion and thus theeffective reproductions rate. For related approaches see [2], [13], [14] [15].
This view of ’lock-down’ corresponds closer to the measures adopted by manygovernments during this second wave. The matching model might have beenmore appropriate during the first wave when indeed in some countries large
79
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 74
-89
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
parts of the entire population were forbidden to leave their home, except foressential business.
Finally we note that social distancing measures (NPIs) cannot affect thenumber of infected, only the rate at which their number grows over time. Eq. 4implies that the only way to avoid all contagion is to completely shut down theeconomy. The parameter ce represents a scaling factor, which would depend onthe structure of the economy (importance of services requiring close contact, liketourism) and the degree to which the population effectively adheres to officialrestrictions. It hast been estimated that ce is of the order of 0.25 [2].
3.1 Uniform control
We first briefly examine the implications of a policy which keeps the number ofinfected [5] and thus the medical load constant. Such a policy is of course notoptimal, but it serves as a useful benchmark for our more general results. Itcan be considered as the limiting case of the ’traffic light’ system in which thedifference in parameters between tiers or levels becomes very small.
Formally, uniform control implies a constant fraction I → Iconst of infected.This is achieved for g = 1, independent of the value of Iconst. The economiccost Econst per time unit is therefore
Econst = ceg0 − gg0
∣∣∣g=1
= ceg0 − 1
g0, (5)
Note that Econst is independent of the value of the medical load one wants toretain. As already mentioned above, the economic costs of keeping the repro-duction factor at one is independent of how many infected there are.
3.2 Stop and go control
Stop and go, or ’bang bang’, control corresponds to an on-off policy as illustratedin Figure 2 above, which can be described within the framework developed hereby the following rule: Control is increased when I = I(t) reaches an upperthreshold Ilarge, and decreased when I = I(t) falls below a lower threshold Ilow.
We denote with gdown < 1 the small reproduction rate corresponding tostrong control, and with gup>1 the large reproduction factor corresponding toweak control. Using the low-incidence approximation (3) we find
Tup =τ
gup − 1ln
(IlargeIlow
), Tdown =
τ
gdown − 1ln
(IlowIlarge
)(6)
for the time Tup needed for I(t) to grow from Ilow to Ilarge, with a respectiveexpression for the down-time Tdown. On a per time basis the economic costs arethen
Ebang =ceg0
[(g0 − gup)Tup + (g0 − gdown)Tdown
] 1
Tup + Tdown(7)
for band-bang, viz stop-and-go control.
80
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 74
-89
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
3.3 Vanishing cost differential
The cost difference between bang-bang and constant control is
Ebang − Econst =ceg0
[1− gupTup + gdownTdown
Tup + Tdown
]. (8)
Note that ln(Ilarge/Ilow) = − ln(Ilow/Ilarge), which implies that both τ andln(Ilarge/Ilow) drop out of (8), which vanishes as
Ebang − Econst =ceg0
[1− gup(gdown − 1)− gdown(gup − 1)
gdown − gup
]≡ 0 . (9)
This implies that both control types, constant and stop-and-go control, comewith the same economic costs.
3.4 Neutrality theorem
The result, that the cost differential between constant and stop-and-go controlvanishes, can be generalised if we rewrite (3) as
τ Ilog = g − 1, Ilog = ln(I) , (10)
where g = g(t) is now an arbitrary function of time. We then have
Ilog(tend)− Ilog(tstart) =
∫ tend
tstart
Ilogdt =
∫ tend
tstart
g(t)− 1
τdt , (11)
which proves that1
tstart − tend
∫ tend
tstart
g(t)dt = 1 (12)
for closed trajectories, viz when Ilog(tend) = Ilog(tstart). Comparing with (5)shows that average economic costs are independent of which timeline g(t) is usedfor controlling the epidemic. Note that the case of constant control, g(t) ≡ g,is included as a special case. If the economic costs of control are proportionalto the reduction in the reproductions rate, one finds thus a ’neutrality theorem’for epidemic control. All trajectories which return to the point of departure (interms of the infection rate) will lead to the same economic cost.
4 Mean number of infected under different con-trol policies
The number of infected becomes the key criterion if the economic cost of differentcontrol policies (over complete cycles) is the same.
81
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 74
-89
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
For constant infection rates g, the cumulative number of infected betweentwo times t = t0 and t = t1 is
X0,1 =
∫ t1
t0
I(t)dt = I(t0)
∫ t1
t0
e(t−t0)(g−1)/τ dt
=I0τ
g − 1
(e(t1−t0)(g−1)/τ − 1
)=
(I1 − I0)τ
g − 1(13)
when using (3) and that I1 = I0 exp((t1 − t0)(g − 1)/τ). Noting that (6) holdsgenerally for constant g, we have
∆T0,1 = t1 − t0 =τ
g − 1ln
(I1I0
), (14)
which leads toX0,1
∆T0,1=
I1 − I0ln(I1/I0)
(15)
for the overall number of infected, on the average per time unit. Note that
ln
(I1I0
)= ln
(I1 − I0 + I0
I0
)≈ I1 − I0
I0(16)
for I1 ≈ I0, from which the limit
limI0→I1
X0,1
∆T0,1= I0 (17)
is recovered. Empirically it been observed that the cumulative medical loadduring the ’down phase’ is about 30% higher than the cumulative load whichfollows the peak [16].
4.1 Bang-bang infection numbers
For stop-and-go control we have two periods with constant g, when I goes upand respectively down. With (15) we find that the total cumulative fraction ofinfected is determined by;
Xbang =Ilarge − Ilow
ln(Ilarge/Ilow)Tup +
Ilow − Ilargeln(Ilow/Ilarge)
Tdown (18)
for bang-bang control, and hence
Ibang =Xbang
Tup + Tdown=
Ilarge − Ilowln(Ilarge/Ilow)
(19)
for the time-averaged number Ibang of infections.Here we are interested in on-the-average stationary control strategies, for
which the level I of infections is kept at the average. In Sect. 3.4 we did show
82
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 74
-89
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
2 4 6 8 10
x = Ilarge
/ Ilow
0.6
0.8
1
1.2
1.4
rati
o m
ean
cas
e n
um
ber
s
(stop and go) / (geometric mean)
(stop and go) / (arithmetic mean)
Figure 3: Relative case numbers for distinct control policies. For stop-and-go control case numbers oscillate between Ilow and Ilarge. Alternatively con-sidered are constant incidences, either at the arithmetic mean, (Ilow + Ilarge)/2,or at the geometric mean,
√IlowIlarge. Shown are the ratios of the respective
per time case numbers, as given by Eqs. (20) and (21).
that stationary control results always in identical economic costs. This holdshowever not for the medical load, which can be considered to be the proportionalto the average infection number I. It is in particular of interest to compare themedical Ibang, of stop-and-go control, with the policies keeping I at a constant,intermediate level or benchmark. A natural choice for this benchmark couldthe arithmetic mean, or midpoint, Imid = (Ilarge + Ilow)/2. However, we willconsider as well the geometric mean Igeo =
√IlargeIlow.
For the arithmetic mean, the ratio Ibang/Imid in mean infection numbers is
δI = Ibang/Imid =
Xbang
Tup + Tdown
2
Ilarge + Ilow
=x− 1
ln(x)
2
x+ 1≤ 1 , (20)
when denoting x = Ilarge/Ilow. In the limes Ilarge → Ilow, viz x → 1, one haslimx→1 δI = 1, as expected. The limes x→ 1 is performed using the small x−1expansion ln(x) = ln(1 + (x− 1)) ≈ x− 1.
That Ibang/Imid is strictly smaller than unity for x > 1 can be seen, e.g.,by plotting (20) as a function of x, as done in Figure 3. Numerically one findsa reduction of 29% at x = 10, which is somewhat higher than ratio of 7 to 1observed in the actual values displayed in Fig. 1. Note that one has 2/(x+1) ≤ 1when x ≥ 1, for the second term in (20), with (x− 1)/ ln(x) ≤ 1 holding for thefirst term. The last relation holds because the log-function is concave, with aslope d ln(x)/dx = 1 at x = 1.
The result that Ibang ≤ Imid implies that a policy of stop and go is superiorto (less bad than) a policy of keeping infections constant half-way between the
83
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 74
-89
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
peak and trough. The two policies would have the same economic cost, butstop and go would lead to a lower overall medical load. Neither policy would ofcourse be optimal in an unconstrained policy space. But our aim is merely toconsider policies that have been adopted. The intuition behind this result canbe seen from Figure 2 above. The infection curve lies more time below thanabove the arithmetic average, a well-known property of exponential growth.
The constant control policy serves only as benchmark. The results are muchmore general: As can be seen from (20) the average medical load falls as the ratioof the upper limit to the lower limit increases, while holding the (arithmetic)average constant. This implies that the medical load resulting from a ’stop andgo’ policy improves as one increases the upper and reduces the lower limit bythe same amount.
Regarding the comparison to the geometric mean Igeo, we consider
Ibang/Igeo =
Ilarge − Ilowln(Ilarge/Ilow)
1√IlargeIlow
=x− 1
ln(x)
1√x≥ 1 , (21)
where we used x = Ilarge/Ilow, as for (20). The functional dependence is includedin Figure 3. It can be seen that it is favorable to keep the incidence at thegeometric mean, instead of letting it oscillate between Ilow and Ilarge. Moreover,the ratio of the respective average medical loads increases only modestly as theratio between upper and lower limit increases. Letting case number vary by afactor of ten leads to an increase of 24%. It should not be surprising that keepinginfections at the geometric mean is worse, given that the distance between thetwo means increases as the ratio of the upper to the lower limit becomes larger.For 9 = Ilarge/Ilow, the arithmetic mean is equal to 5 whereas the geometricmean is 3. Choosing the geometric as the intermediate point thus amounts tochoosing a higher benchmark.
From a general viewpoint, if follows from (21) that the average medicalload increases as the ratio of the upper limit to the lower limit increases, whenholding the geometric average constant. This implies that the medical loadresulting from a ’stop and go’ policy increases as one increases the upper andreduces the lower limit by the identical factor. These considerations suggestthat the decision regarding how wide apart to set the limits of a ’stop and go’or a regional ’traffic light’ policy depends on whether one wants to keep thearithmetic or the geometric average constant.
4.2 Significance
The results discussed above apply only within the limits of hospital capacity.Once that limit has been reached any further increase in the medical load wouldimply rapidly rising medical and ethical costs. A lower medical load constitutes asufficient criterion for preferring the stop and go policy, given that the respectiveeconomic costs vanish for different control paths as long as the initial infection
84
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 74
-89
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
incidence is reached again. Note that the medical load can be translated intoeconomic costs [15], [17], [18], [19].
Most of the literature focuses on the economic value of the lives lost. How-ever, that might be a mistake [2], as infections with less severe symptoms canalso lead to considerable economic costs. One example of the economic cost ofinfections would be the loss of working time of those infected and with mildsymptoms, when these individual have to self-isolate and cannot work for a cer-tain time. This loss could be calculated as the number of weeks of working timedue to symptoms (and self-isolation needs) and would be equal to a proportionof GDP [2]. To this one would have to add the hospitalisation costs for thosewith stronger symptoms and finally the economic value of lives lost. The lowermedical load implied by a stop and go policy would thus also lead to loweroverall economic costs.
A key difference between the present study and most existing literatureconcerns the kind of policies examined. Here we focus on a comparison betweentwo representative, real-world policies, stop and go vs. constant control. Optimalcontrol, which usually does not result in reversals of restrictions, is in contrastthe goal for the majority of studies published so far. Contributions like [10]and [11] employ the matching framework, which is based on confinement as themain policy instrument and allows for economies of scale in lowering contagionas explained above. In such a framework it is clear that the optimal policywould be to impose tight restrictions from the start until the desired incidenceis attained. For example, [10] finds that “the optimal confinement policy is toimpose a constant rate of lock-down until the suppression of the virus in thepopulation.”.
Other contributions, which do not incorporate economies of scale in contain-ment policies (either because they use constant returns in matching or becauseof a different view of social distancing restrictions) arrive at somewhat differenti-ated conclusions. For example, [15] where containment works like a consumptiontax, find that containment should build up gradually and peak early when thereexists the perspective of a vaccine being discovered.
5 Changing epidemic parameters
The native reproduction factor g0 is, as a matter of principle, an intrinsic prop-erty of the virus. As such it can be measured only at the very start of apandemic, namely when nobody is yet aware of what is happening. However,this reproduction factor can change over time, for example with the season. It iswell known that the danger of contagion is much higher indoors than outdoors.It is in fact nearly impossible to catch the Coronavirus outside [20]. In the caseof influenza virus this change in the reproduction rate leads to the typical ’fluseason’, which starts with the onset of colder weather (late in the year in theNorthern Hemisphere). For the case of the Covid-19 virus other seasonal factorshave also been mentioned, for example fluctuations sin UV light[21].
One thus needs to consider time-dependent g0, which would correspond
85
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 74
-89
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
to the underlying spreading rate in a given societal state, ‘without explicit,government-imposed restrictions’.
We assume that g0 changes right at the top for the case of bang-bang con-
trol. Going up/down we then have g0 = g(up)0 and g0 = g
(down)0 . As a further
simplification we postulate
Tup = Tdown, gup − 1 = 1− gdown , (22)
which is equivalent to gup + gdown = 2. Compare (6).
5.1 Costs for the economy
Given that we assume that Tup = Tdown, the per time economic costs of keepingconstant infection numbers is
Econst =ce2
[g(up)0 − 1
g(up)0
+g(down)0 − 1
g(down)0
]. (23)
For bang-bang control we have instead
Ebang =ce2
[g(up)0 − gupg(up)0
+g(down)0 − gdown
g(down)0
]. (24)
The difference is
Ebang − Econst =ce2
[1− gupg(up)0
+1− gdown
g(down)0
], (25)
which simplifies to
Ebang − Econst =ce2
[1
g(down)0
− 1
g(up)0
] (1− gdown
)︸ ︷︷ ︸> 0
(26)
when using the Tup = Tdown condition that gup− 1 = 1− gdown, see (22). Goingdoing down we restrict, viz gdown < 1 holds.
Bang-bang control is therefore favorable when g(down)0 > g
(up)0 . This result
would support the pattern observed in Europe where during the summer (wheng0 was lower than before) governments eased restrictions, imposing them againduring the fall when g0 increased again. The economic cost of the restrictionscould thus be concentrated during the period when they had a higher yield interms of infections avoided.
6 Conclusions
During the first wave of the Covid-19 pandemic the key concern was to ’flattenthe curve’. Harsh lock-down measures were needed when health systems were
86
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 74
-89
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
overwhelmed by the sudden increase in hospitalizations, many of which requiredintensive care units. The second wave has so far resulted in a somewhat lowermedical load, but it proved nevertheless indispensable to re-introduce some so-cial distancing measures as otherwise the case load would have continued toincrease at a near exponential rate. Countries have taken in this regard dif-ferent approaches. In some, the measures have been just enough to stabilizeinfections. In others, the measures have led to a strong fall in new cases andgovernments plan to lift restrictions soon - which is likely to lead to a renewedincrease.
Several countries introduced, interestingly, regionally graduated systems,which allow regions and cities to oscillate between periods of harsh restric-tions that are triggered when infection numbers surpass certain thresholds, andperiods of lower restrictions which start when numbers have fallen again, nowbelow a given threshold. These kind of quasi-automatized threshold contain-ment policies are graded realization of the stop and go policy examined in thepresent study.
We have shown in the context of our model, that any time path of restrictionswhich returns to the point of departure in terms of the infection rate, a scenariolike to occur within rule-based ’traffic light’ systems, imply the same overalleconomic costs. Our analysis suggests furthermore that stop and go policiesmight not be as costly as it might appear at first sight, at least if comparedto the alternative of keeping the incidence rate constantly the midpoint. Theeconomic cost would be the same, but the overall medical load of stop and goshould be lower. The result is reversed for the alternative of keeping infectionsat the (lower) geometric mean. Compare Figure 2.
Applied to regional ’traffic light’ systems, our results imply that increasingthe upper, and reducing at the same time the lower threshold by the sameabsolute amount ∆I, via Iupper → Iupper + ∆I and Ilower → Ilower −∆I leadsto a lower medical load. The opposite would be true if one were to increase theupper, and at the same time reduce the lower threshold by the same proportion,fI > 1, this time using the rescaling Iupper → IupperfI and Iupper → Iupper/fI .The latter procedure would increase the arithmetic average since the absoluteincrease in the upper limit would be now higher than the fall in the lowerlimit. This implies that any choice regarding the range between the upper andlower thresholds in a regional ’traffic light’ must take this difference betweenarithmetic and geometric mean into account.
We also show that it makes sense for policies to react to seasonal variationsin the native reproduction factor. The economic cost of tight social distancingshould be incurred in winter, when infections would otherwise be high and rising.Bang hard on the virus when Santa Claus knocks at the door.
References
[1] Linda Thunstrom, Stephen C Newbold, David Finnoff, Madison Ashworth,and Jason F Shogren. The benefits and costs of using social distancing
87
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 74
-89
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
to flatten the curve for covid-19. Journal of Benefit-Cost Analysis, pages1–27, 2020.
[2] Claudius Gros, Roser Valenti, Kilian Valenti, and DanielGros. Strategies for controlling the medical and socio-economiccosts of the corona pandemic. https://clausen.berkeley.edu/wp-content/uploads/2020/04/Corona.pdf, 2020.
[3] Rajib Shaw, Yong-kyun Kim, and Jinling Hua. Governance, technology andcitizen behavior in pandemic: Lessons from covid-19 in east asia. Progressin disaster science, page 100090, 2020.
[4] Naoki Yamamoto and Georg Bauer. Apparent difference in fatalities be-tween central europe and east asia due to sars-cov-2 and covid-19: Fourhypotheses for possible explanation. Medical hypotheses, 144:110160, 2020.
[5] Eric B Budish. Maximize utility subject to r ≤ 1: A simple price-theory ap-proach to covid-19 lockdown and reopening policy. NBER Working Paper,(w28093), 2020.
[6] Clauadius Gros, Fabian Schubert, and Carolin Roskothen. Goethe Inter-active Covid-19 Analyser, 2020.
[7] Vincenzo Capasso and Gabriella Serio. A generalization of the kermack-mckendrick deterministic epidemic model. Mathematical Biosciences, 42(1-2):43–61, 1978.
[8] Herbert W Hethcote and P Van den Driessche. Some epidemiological mod-els with nonlinear incidence. Journal of Mathematical Biology, 29(3):271–287, 1991.
[9] JHU-CSSE. John Hopkins Center of Systems Science and EngineeringCOVID-19 repository, 2020.
[10] Christian Gollier. Pandemic economics: optimal dynamic confinement un-der uncertainty and learning. The Geneva Risk and Insurance Review,45(2):80–93, 2020.
[11] Daron Acemoglu, Victor Chernozhukov, Ivan Werning, and Michael DWhinston. A multi-risk sir model with optimally targeted lockdown. Tech-nical report, National Bureau of Economic Research, 2020.
[12] Robert Rowthorn and Jan Maciejowski. A cost–benefit analysis ofthe covid-19 disease. Oxford Review of Economic Policy, 36(Supple-ment 1):S38–S55, 2020.
[13] Aaron Strong and Jonathan Welburn. An Estimation of the EconomicCosts of Social-Distancing Policies. 01 2020.
[14] Dirk Krueger, Harald Uhlig, and Taojun Xie. Macroeconomic dynamicsand reallocation in an epidemic. Covid Economics, 5:21–55, 2020.
88
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 74
-89
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
[15] Martin Eichenbaum, Sergio Rebelo, and Mathias Trabandt. The macroe-conomics of epidemics. NBER Working Paper No. 26882, 2020.
[16] Claudius Gros, Roser Valenti, Lukas Schneider, Benedikt Gutsche, andDimitrije Markovic. Predicting the cumulative medical load of covid-19outbreaks after the peak in daily fatalities. medRxiv, 2020.
[17] Andrew Atkeson. What will be the economic impact of covid-19 in the us?rough estimates of disease scenarios. NBER Working Paper No. 26867,2020.
[18] Abel Brodeur, David Gray, Anik Islam Suraiya, and Jabeen Bhuiyan. ALiterature Review of the Economics of COVID-19. IZA DP No. 13411,IZA Institute of Labour Economics, 2020.
[19] Richard Baldwin. Keeping the lights on: Economic medicine for a medicalshock. Macroeconomics, 20:20, 2020.
[20] Ivan Couronne. Catching coronavirus outside is rare but not impossible.MedicalXpress, 2020.
[21] Princeton University Bendheim Center for Finance. Bengt Holmstrom:The Seasonality of Covid-19, 2020.
89
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 74
-89
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
Covid Economics Issue 62, 18 December 2020
Copyright: So Kubota, Koichiro Onishi and Yuta Toyama
Consumption responses to COVID-19 payments: Evidence from a natural experiment and bank account data1
So Kubota,2 Koichiro Onishi,3 and Yuta Toyama4
Date submitted: 9 December 2020; Date accepted: 14 December 2020
We document households' spending responses to a stimulus payment in Japan during the COVID-19 pandemic. The Japanese Government launched a universal cash entitlement program offering a sizable lump sum of money to all residents to alleviate the financial burden of the pandemic on households. The timings of cash deposits varied substantially across households due to unexpected delays in administrative procedures. Using a unique panel of 2.8 million bank accounts, we find an immediate jump in spending during the week of payments, followed by moderately elevated levels of spending that persist for more than a month. The implied marginal propensity to consume is 0.49 within 6 weeks. We also document sizable heterogeneity in consumption responses by recipients' financial status and demographic characteristics. In particular, liquid asset holdings play a more crucial role than total asset holdings, suggesting the importance of the wealthy hand-to-mouth.
1 The data used in this study were obtained under an academic agreement between Mizuho Bank and Waseda University. The views and opinions expressed in this paper are solely those of the authors and do not reflectthose of Mizuho Bank. We are extremely grateful to the staff at Mizuho Bank for answering numerous questions regarding the data. Administrative and financial support by the Center for Data Science at Waseda University is greatly appreciated. Yuji Mizushima and Motoki Murakami provided superb research assistance. We also thank Satoshi Tanaka, Takashi Unayama, and the seminar participants at the Bank of Japan and The 22st Macroeconomics Conference for their comments and advice. All errors are our own.
2 Associate Professor, School of Political Science and Economics, Waseda University3 Associate Professor, School of Education, Waseda University4 Associate Professor, School of Political Science and Economics, Waseda University
90
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 9
0-12
3
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
1 Introduction
The COVID-19 pandemic and the subsequent lockdowns had severe impacts on household budgets.
A large number of studies have documented drastic declines in income, spending, and debt pay-
ments in various countries, including the United States (Baker et al., 2020a; Chetty et al., 2020;
Coibion et al., 2020a; Cox et al., 2020), the United Kingdom (Hacioglu et al., 2020; Carvalho et al.,
2020), Spain (Garcıa-Montalvo and Reynal-Querol, 2020), Sweden and Denmark (Sheridan et al.,
2020), and Japan (Watanabe, 2020). Moreover, the pandemic shock disproportionately a↵ected
groups with certain socioeconomic backgrounds and from di↵erent sectors; for instance, those in
the service industry, those in occupations that cannot be performed at home, low socioeconomic
status households, minorities, women, youth, and parents were particularly susceptible (See, e.g.,
Forsythe et al. (2020); Alon et al. (2020); Montenovo et al. (2020) for the US, Blundell et al. (2020)
for the UK, Kikuchi et al. (2020) for Japan, and Adams-Prassl et al. (2020); Belot et al. (2020) for
cross-country studies).
To mitigate economic shocks, governments have enacted urgent relief measures, primarily in
the form of cash payments. Most OECD countries have introduced new systems for household
cash transfers (OECD, 2020). These programs’ extraordinary budgets and legislators’ inclinations
toward further stimulus call for a serious quantitative evaluation of COVID-19 cash transfer policies.
These quantitative evaluations also provide implications for key macroeconomic variables, such as
households’ marginal propensities to consume (MPCs) and fiscal multipliers.
In this paper, we examine the Special Fixed Benefits program (hereafter, SFB), a large-scale
cash-transfer program launched by the Japanese government in response to the COVID-19 pan-
demic. The program entails a fixed and sizable cash transfer amounting to 100,000 JPY (approx-
imately 950 USD) to every individual in Japan who applies regardless of age, income, family size,
and employment. The payment policy provides a “natural experiment” because the timing of pay-
91
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 9
0-12
3
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
ments was rendered nearly random due to the huge overcapacity in the administrative procedures
at local o�ces. Our data reveal a continuous and bell-shaped distribution of payment days be-
tween the second week of May and the end of August. Variations in timing of payments allows us
to estimate precisely the immediate e↵ect of SFB payments on household spending within a brief
temporal window. Our event study approach lets us separate the e↵ects of SFB payments from the
e↵ects of pandemic shocks, stay-at-home measures, and other policies.
We examine high-frequency transaction-level data for 2.8 million personal accounts at Mizuho
Bank, one of Japan’s three largest commercial banks, that recorded SFB payments during 2020.
The dataset is a panel of bank accounts, containing information about the account balances, the
inflows to and outflows from the accounts, and basic demographic information about the account
holders. We explore the heterogeneous e↵ects on consumption of such demographic characteristics
as account holders’ income, income loss attributable to the pandemic, total assets, demand deposit
balances, and liquidity constraints. Our main indicator of spending is total outflows, including
ATM withdrawals, transfers to other bank accounts, and credit card charges. As a result, our
estimates of MPC constitute an upper bound on household spending.1
Our main findings imply that the MPC is 0.49 within six weeks of receipt among all samples.
Our estimates are comparable to those obtained from studies of the U.S. Coronavirus Aid, Relief,
and Economic Security (CARES) Act.2 Among dynamic response, we find an immediate jump in
spending during the week of payments and thereafter moderate increase that persists more than
a month.3 ATM withdrawals constitute 63% of the total outflows, a reasonable percentage given
1Evidence from the US COVID-19 stimulus package suggests this upper bound reasonably captures actual spend-ing. Small di↵erences appear between MPCs estimated by a consumption survey (Coibion et al., 2020b) and by totaloutflows from bank accounts (Karger and Rajan, 2020).
2Karger and Rajan (2020) and Misra et al. (2020) find MPCs of 0.50 and 0.43, respectively, using the financialtransaction data from Facteus.
3The sustained increase in expenditure may be attributable to our large sample size. Previous evidence fromJapan is unstable to specifications and data. Shimizutani (2006) analyzes the 1998 tax cut and reports an MPC of0.6 during its first month of implementation, but its e↵ects nearly dissipate as shown by negative coe�cients duringthe second month. Koga and Matsumura (2020) leverage a subjective question in the Survey of Household Financeand elicit a self-reported MPC of 0.73. However, they note a significant bias in self-reported data and find a more
92
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 9
0-12
3
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
that cash is the major payment method in Japan (Fujiki and Tanaka, 2018).
We observe heterogeneity in MPC across recipients’ financial status and demographic charac-
teristics. One substantial variation is caused by credit constraints. Recipients with binding credit
constraints spend more (59%) of their SFB transfers. MPC hinges on the liquidity of assets. We
find significant variation in MPC by quartile of demand deposits, one of the most liquid assets.
However, total asset holdings make almost no di↵erence in MPC. We also find a large MPC among
wealthy hand-to-mouth, who hold a small demand deposit balances but extensive total assets. This
result implies the crucial roles of the distributions over both liquid and illiquid asset holdings in
fiscal and monetary policy analyses (Kaplan and Violante, 2014; Kaplan et al., 2018). Analysis of
household income reveals a modestly large MPC (0.54) among households that su↵ered COVID-
related losses exceeding 50% of their 2019 income. On the other hand, there is little variation in
MPC across income quartiles. We also find that SFB exhibits a uniform e↵ect across family size.
Taken together, our results suggest both the potential improvements and limitations of targeting
transfers when designing programs to stimulate consumption.4
Related Literature Our study extends a growing literature on the estimates of the MPCs from
COVID-19 stimulus payments. One strand of the literature examines the US CARES Act. Baker
et al. (2020b) use financial transaction data from fintech app users to derive an MPC spanning
0.25 to 0.35. Respondents to a survey by Coibion et al. (2020b) self-report an average MPC
of approximately 0.4, although MPC varies considerably across respondents’ attributes such as
homeownership and liquidity constraints. Karger and Rajan (2020) use transaction-level data for
conservative MPC of 0.16 by tracking transitory income shocks in the Japan Household Panel Survey. Examininganother type of cash transfer, Hsieh et al. (2010) found an MPC of 0.1–0.2 for a shopping coupon policy implementedin 1999. Macroeconomists have supposed the recent Japanese MPC values to be low because of the small fiscalmultipliers estimated from time-series analyses; for example, see Auerbach and Gorodnichenko (2017). Employing aquantitative macroeconomic model, Braun and Ikeda (2020) emphasize the role of Japanese SFB policy during thecurrent pandemic in mitigating consumption inequality.
4Indeed, stimulating aggregate consumption does not directly lead to welfare improvement. Nygaard et al. (2020)theoretically studies optimal policy design and find that the CARES Act would have improved social welfare byredistributing payments toward low-income and young recipients.
93
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 9
0-12
3
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
debit cards from Facteus and find an average MPC of 0.5. They also find considerable heterogeneity
in consumption responses that can be partially attributed to observable characteristics such as age,
income, and location. Using regional variations in the timing of stimulus payments and Zip code
data from Facteus, Misra et al. (2020) find an average MPC of 0.43 during the initial four days upon
receiving a payment and decompose their results by product categories. Chetty et al. (2020) build
a real-time zip code-level database of socio-economic variables obtained from private companies.
They find significant jumps in consumer spending and business revenue around mid-April, which
overlaps the period of payments from CARES Act. Previous studies also estimate MPCs from the
2001 Economic Growth and Tax Relief Reconciliation Act and the 2008 Economic Stimulus Act
in the US (Shapiro and Slemrod, 2003; Johnson et al., 2006; Agarwal et al., 2007; Shapiro and
Slemrod, 2009; Parker et al., 2013; Broda and Parker, 2014; Misra and Surico, 2014).
A few studies estimate MPCs outside the United States. Liu et al. (2020) study a temporary
small-scale digital discount coupon intended to aid Chinese consumers during the COVID-19 crisis.
Using transaction-level data from Alipay e-wallet, they find strikingly high MPC of 3.4 to 5.8,
which they attribute to consumers’ behavioral reactions. Kim and Lee (2020) examine a household
consumption voucher initiative in South Korea, and find that most survey respondents planned to
spend more on necessities.
Our study o↵ers multiple contributions to the literature concerning stimulus payments and
MPC, especially as pertains to the COVID-19 pandemic. First, we exploit the random variation in
timings of payments in an event study design. Second, our results are easily interpreted due to the
Japanese government’s fixed-payment scheme. Third, our use of high-frequency bank account data
featuring millions of observations lets us derive precise estimates. Fourth, our results accurately
reflect personal MPCs because we observe the date of SFB payments.
This study also ties to empirical studies that use bank account data to investigate the economic
94
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 9
0-12
3
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
e↵ects of the COVID-19 pandemic. Farrell et al. (2020) examine the e↵ects of US Unemployment
insurance benefits using Chase bank account data that shares characteristics with our data. The
estimated MPC of unemployment insurance benefits (0.73) is reasonably comparable with our
results for financially constrained households. Sheridan et al. (2020) use data from Danske Bank in
Scandinavia to compare individual spending responses in Denmark, which imposed strict shutdown
policies, and Sweden, which imposed moderate stay-at-home orders. Their results imply that social
distancing causes only minor decreases in spending. Carvalho et al. (2020) use transaction-level
data from the BBVA in Spain to document changes in consumption during the pandemic and
lockdown. Bounie et al. (2020) document a sharp rise in wealth inequality during the pandemic
using randomly selected accounts at a French bank, Credit Mutuel Alliance Federale.
2 The COVID-19 Pandemic and the Special Fixed Benefits Policy
in Japan
2.1 COVID-19 Crisis in Japan
The first COVID-19 case was reported on January 6, 2020, which is relatively early compared
with the timeline of the first cases in other countries. The number of reported cases began to
increase after January 28 as Japan faced shortages of masks and hand sanitizers. Several weeks
into the pandemic, an outbreak occurred on the Diamond Princess cruise ship that resulted in 712
infections and 13 deaths. The spread of COVID-19 was substantially slower in Japan than in many
other countries, but the number of cases in Japan still exhibited a pattern of exponential growth.
To combat the rise in infections, the Japanese Government mandated a temporary closure of all
elementary, middle, and high schools on February 27. On March 24, Japan postponed the Tokyo
2020 Summer Olympics and Paralympics.
95
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 9
0-12
3
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
The number of cases rose rapidly after late March, given Japan’s weak surveillance and limited
capacity for PCR testing. The Japanese Government announced its first state of emergency on
April 7 for urban areas, and extended it nationwide on April 16. Unlike the stringent lockdown
policies of other countries, Japan’s announcement lacked legal binding force. Nevertheless, the de-
facto stay-at-home order reduced outings 20% (Watanabe and Yabu, 2020). The first-wave of the
pandemic peaked around mid-April and was nearly contained by mid-May. The state of emergency
was lifted selectively on May 14 and everywhere on May 25. Aggregate damage to public health in
Japan was relatively minor: by May 25, Japan had reported 16,706 cases and 846 deaths among
its 126.5 million people.
Economic damage was substantial, however. Compared to consumption in April and May of
2019, consumption in 2020 was 11.1% lower in April and 16.2% lower in May 2020, according to the
Family Income and Expenditure Survey. The unemployment rate changed little because of Japan’s
entrenched employment protection, according to the labor force survey, but hours worked fell 3.9%
in April and 9.3% in May. Wages of full-time workers dropped 0.7% in April and 2.8% in May.
2.2 The Special Fixed Benefit Payment
We evaluate responses of household expenditures to the SFB policy, the largest COVID-19 relief
program in Japan’s first supplemental-spending bill.5 The program entitled all Japanese residents
to a one-time payment without restrictions on age, income, family size, or nationality. The amount
allocated per person was 100,000 Japanese JPY (approximately 950 USD), about 42% of the median
monthly earned income of a full-time worker.
All Japanese households were notified by mail, and asked to apply online or by mail. A head of
households applied for benefits for all family members in the same residence. Application requires
5Ando et al. (2020) summarize Japan’s COVID-19 relief programs.
96
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 9
0-12
3
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
individual identification numbers of all family members and bank account information.6 In house-
holds with more than one resident, all benefits were deposited into the head of household’s bank
account.
The SFB payment is unique and it is ideal for studying consumer responses to a one-time fiscal
stimulus for two reasons. The SFB was the only universal, fixed-sum COVID-19 relief payment
among advanced economies. According to OECD (2020), all other OECD countries imposed condi-
tions on transfers (e.g., income eligibility thresholds or enrollment in social security).7 Second, and
more importantly, the payment date was nearly random within a range of several weeks because
of administrative constraints. Local governments had to check each household’s application and
send remittances to bank accounts manually.8 Applications were primarily by mail because online
application failed following technical di�culties. This led to an overflow of mailed applications and
subsequent administrative errors. When such errors occurred, local o�ces had to correspond with
applicants by phone or email to correct them. Some municipalities denied applications with incom-
plete entries or errors. For instance, 20% of applications in Saga city, a middle-sized municipality,
had errors.9
News reports suggest that payment dates varied, depending on cities’ administrative capacities
and the experience of o�ce sta↵. Among large cities, Sapporo city had nearly completed remit-
tances to all applicants by mid-June, Nagoya city had not even finished sending applications to
6Although applying was not mandatory, nearly all Japanese households applied. For instance, 98% of residents inYokohama had applied for SFB by August 31, 2020.City of Yokohama webpage: https://www.city.yokohama.lg.jp/lang/covid-19-en/fixed-sum.html.
7South Korea’s universal COVID-19 financial relief program issued vouchers (Kim and Lee (2020)), but, theamount was small ($83 to $332 USD per person) and depended on household size and municipality of residence.Vouchers could be used only within a household’s region of residence, except for bars, clubs, online stores, and largeretailers. The CARES Act in the US was the only relief program comparable to Japan’s SFB payments (Baker et al.,2020b; Chetty et al., 2020; Coibion et al., 2020b; Li et al., 2020; Misra et al., 2020). Amounts provided by CARESwere large enough to detect consumer responses to the stimulus, but they were means-tested: a single person earningbelow 75,000 USD was entitled to the full amount (1,200 USD), but higher-income households received substantiallyless. A similar income-payment scheme was applied to married couples. Each child was eligible for 500 USD.
8Individual identification numbers are not linked to bank accounts in Japan, unlike elsewhere (e.g., the US).9Saga TV, 20% of Application Documents for 100,000 JPY Payment are Incomplete in Saga City, Saga Prefecture
(in Japanese), May 21, 2020.https://www.sagatv.co.jp/news/archives/2020052102735
97
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 9
0-12
3
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
residents.10 In Tokyo Prefecture, 85% of applicants living in Nerima Ward received payments by
June 30, whereas only 34.5% of residents in Edogawa Ward had received payments by then.11 Even
applicants from the same municipality received payments at di↵erent time. Applications submitted
on the same day could result in di↵erences in the timing of payments by several days.12 One error
on an application could delay a payment by more than two weeks.13 Such lack of uniformity in
time-to-payment was unique among COVID-19 relief programs worldwide. For example, time-to-
payment in the US di↵ered only by whether payment was by direct deposit or paper check (Baker
et al. (2020b)).
3 Data and Descriptive Analysis
3.1 Data
Our data are account-level daily transactions from more than 24 million accounts at Mizuho Bank
spanning January 2019 to August 2020. Data include transaction dates, payments and withdrawals,
remarks about each transaction, and end-of-month balances. Data also identify some specific
transactions (e.g., salary payments and ATM withdrawals). We also have demographic information
such as age, gender, and municipality level address. Data were anonymized for all account holders.
To examine how the SFB payment a↵ected household consumption behavior during the COVID-
19 crisis, we restrict our sample to accounts that had received the payment by August 31. We
identify those accounts through transaction remarks and deposit amount.14 The resulting sample
of 2.8 million head-of-household accounts constitutes 4.8% of all Japanese households. Although
10Nikkei news, “Significant delay in 100,000 JPY benefit in Nagoya city: long time for system maintenance” (inJapanese), June 26, 2020.
11J-CAST news, What we ask the Governor of Tokyo is “Quick transfer of 100,000 yen.” (in Japanese), July 6,2020.
12Some municipalities, including Edogawa Ward, announced such case in their web page.13Saga TV, May 21, 2020.14We identify SFB payments in two way: by remarks on accounts indicating an SFB payment (Teigaku or Kyufu
in Japanese) and by deposits that were exactly multiples of 100,000 JPY.
98
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 9
0-12
3
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
Table 1: Summary Statistics
Panel A: Account level variables N Mean St. Dev. 25% Median 75%
SFB Payment (JPY) 2,832,537 204,125 118,615 100,000 200,000 300,000Week of Deposit 2,832,537 26.913 2.778 25 27 29Age 2,804,678 53.065 17.707 39.000 52.000 66.000Female dummy 2,809,140 0.258 0.438 0.000 0.000 1.000Wealth (JPY) 2,809,140 4,200,102 17,019,527 127,000 650,000 3,440,000Demand-deposit balance (JPY) 2,809,140 2,721,917 11,549,024 94,000 444,000 2,092,000Monthly Salary in 2019 (JPY) 1,419,299 276,653 392,953 169,328 250,446 342,939Monthly Salary in 2020 (JPY) 1,419,299 272,650 374,138 166,342 245,522 338,404COVID-19 Shock Dummy 1 1,419,299 0.129 0.335 0.000 0.000 0.000COVID-19 Shock Dummy 2 1,419,299 0.097 0.296 0.000 0.000 0.000Liquidity Constraint Dummy 1,201,443 0.283 0.450 0.000 0.000 1.000
Panel B: Account-week level variables N Mean St. Dev. 25 % Median 75%
Total Withdrawal in 2020 (JPY) 99,138,795 132,855 2,476,128 0 19,569 95,860Cash Withdrawal in 2020 (JPY) 99,138,795 31,885 134,501 0 0 21,000
Notes: COVID-19 shock dummy 1 and 2 represent 15-50% and more than 50% decline in monthly salary in April and May
2020, respectively. Liquidity constraint dummy takes 1 if an end-of-month account balance was below account holders’ monthly
income.
our sample size is abundant and rich in data, it represents only accounts at Mizuho Bank. Mizuho
has branches throughout Japan, but accounts concentrate in larger cities, especially around Tokyo.
Therefore, our analysis may reflect the behavior of urban recipients of the SFB payment.
3.2 Descriptive Statistics
Table 1 presents summary statistics of our dataset. We drop the minimums and the maximums of
all variables to maintain anonymity. Panel A summarises account-level variables. The average SFB
payment is approximately 200,000 JPY. This is because payments are fixed at 100,000 JPY per
person, and there are, on average, two family members residing in each household of our sample.
Their average wealth in April 2020 was approximately 4.2 million JPY; demand-deposit balance (2.7
million JPY) comprised 65% of wealth. Table 1 reports that head-of-household income averaged
270,000 JPY in April and May 2019 and 2020. It had slightly declined between 2019 and 2020.
99
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 9
0-12
3
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
Note that monthly salary is customarily deposited in employees’ bank accounts in Japan.15
We construct two variables to measure financial distress from the COVID-19 crisis. The first
variable (COVID-19 Shock Dummy 1) is a dummy that indicates whether account holders experi-
enced a 15-50% declines in monthly salary in April and May 2020 relative to those months of 2019.
This group includes employees placed on temporary leave.16 Leave allowances usually equal 60%
to 80% of salaries. The second variable (COVID-19 Shock Dummy 2) indicates declines in monthly
earned income exceeding 50%. As Japan’s unemployment rate remained nearly unchanged, this
variable captures self-employed workers whom the COVID-19 crisis damaged considerably. The
Japanese government assists small business whose revenues fell more than 50% from pre-crisis lev-
els. Table 1 indicates that 13% of households experienced 15–50% declines in income and that 10%
experienced declines exceeding 50% from the same months of the previous year.
Our dummy to measure liquidity constraints takes a value of 1 if an end-of-month account
balance was below account holders’ monthly income. We define “the end of the month” as the
month preceding to the account’s SFB payment. Results show that about 28% of accounts are
liquidity constrained.
Panel B of Table 1 presents summary statistics of weekly account total and cash withdrawals
as time-variant dependent variables of interest. Withdrawals include remittances to other bank
accounts, for utilities and rent, and credit card payments. The mean (median) of weekly cash
withdrawals is 132,855 JPY (19,569 JPY). The mean (median) of weekly cash withdrawals is
31,885 JPY (0 JPY), indicating no cash was withdrawn from at least half of sampled accounts in
any given week.
15This salary likely reflects the head of household’s monthly disposable income because tax and social insuranceusually are deducted monthly in Japan.
16Granting temporary leave has been Japanese companies’ dominant response to COVID-19 because layo↵s aredi�cult under Japan’s employment protection. In response to the economic recession related to COVID-19, JapaneseGovernment has proposed Employment Adjustment Subsidies for paying leave allowance.
100
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 9
0-12
3
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
Figure 1: Timing of Deposits of SFB Payments
3.3 Timing of the SFB Payment
Figure 1 is a histogram of numbers of SFB payments to households over the sampled period. The
majority of transfers appear between late June and early July, the earliest appears in May, and
the last appears during the final week of August. As Section 2.2 discussed, the administrative
delays made the timing of payments unpredictable. We provide further support for this claim by
regressing the week of households’ SFB recipience on various demographic variables and geographic
indicators in Appendix A.
4 Empirical Strategy
To estimate the e↵ect of SFB payments on household spending, we leverage weekly variations in
timing of SFB payments across households in the following event study specification:
yitw = ↵i + ↵iw + ↵tw +bX
k=a
�kDkitw + uitw, (1)
101
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 9
0-12
3
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
where yitw is an outcome measure for bank account i in week w(= 1, . . . , 35) in year t(= 2019, 2020).
We use both cash and total withdrawal per capita as a measure of household spending.17 ↵i
denote account-level fixed e↵ects that capture time-invariant heterogeneity across households. ↵iw
is account-by-week fixed e↵ects that control for seasonal patterns of consumption that are specific
to each household. For instance, families with small children increase consumption significantly
more than single-member households around Children’s day in early May. ↵tw are the year-by-
week fixed e↵ects that capture aggregate shocks and national policies. Later, we allow these time
fixed e↵ects to vary across regions (i.e., prefectures) to account for the potentially heterogeneous
economic e↵ects of COVID-19 across prefectures. uitw is idiosyncratic error.
The independent variable of interest is Dkitw, where Dk
itw = 1 {w � Ti = k}, and Ti denotes the
week in which account i receives an SFB payment. Let k 2 [a, b] be the event-time relative to the
week households receive SFB payments. The week prior to the deposit corresponds to k = �1, and
the week of payment is given by k = 0. We set a = �5 and b = 5 in our analysis. Coe�cient �k
(for k >= 0) captures household spending responses k-weeks after deposit of SFB payments. We
also include the lead terms (for k < 0) to test for the presence of the pre-trends in the k weeks
preceding the payment. We normalize the coe�cient ��1 to 0.
Estimating Eq. (1) directly using a within transformation is computationally intractable be-
cause of the enormous sample size, which amounts to approximately 200 million weekly-account
cells spanning two years of transactions. Therefore, we begin by computing di↵erences across ob-
servational years to eliminate fixed e↵ects ↵i and ↵iw from Eq. (1). The resulting specification is
�yiw = �↵w +bX
k=a
�kDkiw +�uiw, (2)
17We are able to infer household size from amounts of SFB payments because each person receives 100,000 JPY.
102
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 9
0-12
3
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
where �xiw = xi,2020,w � xi,2019,w denotes changes in variable x from 2019 to 2020 within each
unique bank account and week. We then employ ordinary least squares to estimate Eq. (2) and
cluster the standard error at prefectural level to account for the serial correlation across households
and over time.
5 Estimation Results
5.1 Results on Full Sample
Table 2 reports estimation results of our event-study analysis. Results of full-sample regressions
appear in Columns (1)-(4). The dependent variable in Columns (1) and (2) is total outflows from
accounts in a given week. Columns (3) and (4) examine weekly cash withdrawals as the dependent
variable. Columns (2) and (4) allows the time fixed e↵ects to interact with prefecture fixed e↵ects.
Households immediately increased total withdrawals by approximately 19,000 JPY during the
week they received SFB payments. Although the response is moderate in the weeks after the
deposit, it remains sizable and statistically significant. Columns (3) and (4) indicate that households
withdraw 15,000 JPY in cash during the week of the deposit, suggesting that spending responses are
driven primarily by cash withdrawals. This finding accords with Japanese households’ preference
for cash over credit or debit cards for purchases.
The right and left panels of Figure 2 plot the event study coe�cients �k from Columns (2)
and (4), respectively, in Table 2. The figure confirms a spike in withdrawals upon receiving SFB
payments. Moderate but statistically significant positive coe�cients of withdrawals persisted for
five weeks.
103
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 9
0-12
3
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
Table 2: Results from Event Study Analysis
Dependent Variables
Total Outflows ATM Withdrawal Total Outflows
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16) (17)
5 weeks prior to payment �3, 452 �3, 522 �152 �146 �6, 405 �1, 319 �2, 123 341 �76 �3, 741 4, 953 �220 �5, 038 �309 2, 299 �1, 351 �7, 388
(1, 977) (1, 933) (79) (73) (3, 768) (1, 310) (954) (1, 052) (401) (790) (3, 336) (616) (2, 097) (265) (3, 137) (1, 664) (4, 642)
4 weeks prior to payment 955 879 �106 �135 2, 527 99 �2, 776 �2, 103 �262 �3, 198 2, 786 �332 982 �1, 088 3, 612 �128 2, 812
(1, 486) (1, 502) (67) (67) (2, 587) (1, 431) (978) (938) (704) (1, 455) (2, 158) (688) (1, 262) (323) (2, 890) (651) (3, 285)
3 weeks prior to payment �587 �627 150 106 �1, 105 1, 961 �2, 710 �4, 511 54 �1, 568 2, 025 �579 293 �960 5, 275 �3, 230 �660
(1, 179) (1, 102) (109) (119) (2, 559) (2, 631) (1, 363) (708) (693) (1, 171) (889) (599) (1, 100) (288) (2, 879) (1, 318) (2, 475)
2 weeks prior to payment 2, 648 2, 619 49 8 4, 174 1, 998 �338 �513 1, 041 2, 082 3, 594 �1, 379 1, 324 �184 6, 459 121 5, 307
(1, 054) (1, 128) (92) (99) (2, 217) (1, 881) (1, 328) (492) (789) (1, 042) (3, 063) (527) (1, 766) (176) (3, 550) (1, 328) (2, 341)
Week of payment 19, 190 19, 050 15, 100 15, 096 16, 418 18, 030 21, 281 21, 393 16, 946 18, 906 26, 918 33, 956 13, 978 28, 675 19, 502 9, 384 10, 163
(799) (806) (312) (311) (1, 433) (2, 574) (2, 523) (529) (726) (1, 082) (2, 835) (843) (1, 210) (288) (2, 500) (1, 483) (1, 280)
1 week after payment 11, 708 11, 534 8, 093 8, 126 12, 206 9, 674 9, 914 11, 040 12, 753 9, 032 14, 383 13, 434 11, 267 13, 354 18, 140 8, 606 8, 917
(1, 550) (1, 492) (133) (126) (1, 826) (557) (3, 650) (1, 405) (1, 242) (2, 007) (2, 051) (1, 308) (2, 781) (636) (4, 264) (1, 293) (2, 196)
2 weeks after payment 5, 550 5, 416 3, 291 3, 323 5, 449 6, 028 4, 362 2, 759 6, 993 3, 503 5, 178 5, 206 5, 660 5, 321 �93 2, 964 6, 153
(1, 371) (1, 277) (119) (109) (1, 653) (1, 905) (957) (542) (1, 205) (927) (3, 852) (745) (2, 371) (153) (1, 136) (1, 541) (2, 609)
3 weeks after payment 5, 064 5, 048 2, 022 2, 045 6, 399 2, 343 3, 021 4, 712 2, 370 3, 598 2, 693 2, 604 4, 048 3, 343 �36 3, 218 7, 225
(930) (1, 002) (86) (82) (2, 286) (756) (1, 019) (1, 095) (762) (2, 345) (2, 570) (637) (1, 263) (315) (4, 778) (1, 887) (1, 948)
4 weeks after payment 2, 872 2, 778 1, 481 1, 498 2, 307 3, 497 2, 220 1, 686 2, 661 5, 263 3, 286 1, 194 3, 800 1, 701 5, 424 2, 415 3, 234
(922) (938) (108) (108) (1, 497) (1, 224) (1, 085) (668) (616) (2, 546) (1, 200) (1, 473) (1, 335) (450) (1, 551) (1, 765) (1, 802)
5 weeks after payment 4, 942 4, 825 1, 159 1, 165 5, 797 6, 301 1, 523 3, 445 3, 800 2, 636 1, 678 �19 7, 222 1, 320 270 �1, 647 9, 213
(1, 908) (1, 892) (89) (92) (2, 442) (2, 790) (726) (801) (1, 582) (1, 492) (1, 399) (532) (4, 068) (263) (3, 510) (1, 309) (3, 578)
Family Sizes 1 2 3 4
COVID-19 Income Shocks < 15% 15� 50% > 50%
Liquidity Constraints Yes No
Demand-Deposit Balance Low Low High High
Total Wealth Low High Low High
Week FE Yes Yes
Week*Prefecture FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Sample Size (millions) 98.1 98.1 98.1 98.1 44.3 23.3 15.4 11.8 30.0 5.1 3.1 11.8 30.1 44.0 5.1 5.0 44.0
Notes: The standard errors are in parenthesis and clustered at prefecture level.
104
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 9
0-12
3
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
Figure 2: Responses to SFB Payments: Full Sample
Cash Total
−5 −4 −3 −2 −1 0 1 2 3 4 5 −5 −4 −3 −2 −1 0 1 2 3 4 5
0
10000
20000
Week relative to the timing of fiscal payment
Coe
ffici
ents
Notes : The figure plots the estimated coe�cients �k for k 2 {�5, . . . ,�1, 0, 1, . . . , 5}. Estimatesare from Columns (2) and (4) in Table 2. Note that ��1 is normalized to 0. Bars indicate 95percent confidence intervals. Standard errors are clustered at prefectural level.
5.2 Discussion on Identifying Assumptions
A key identifying assumption in our event-study design is the parallel trend in withdrawal amounts
between households that di↵er in the timing of SFB deposits. Although our assumption of a parallel
trend is not directly testable, we provide supporting evidences.
First, we examine the lead coe�cients on the event study design. Most of the coe�cients before
SFB payments are statistically insignificant (right panel of Figure 2) or are precisely estimated
as zero (left panel). These estimates imply absence of any pre-trend in household consumption,
suggesting that a parallel trend assumption is likely to hold. This finding aligns with our discussion
in Section 3.3 and Appendix A on the plausible exogeneity of payment timings within a narrow
time window.
Second, we investigate the robustness of our results by including week-by-prefecture fixed ef-
105
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 9
0-12
3
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
fects. The concern here is the correlation between the timing of SFB payments and regional
macroeconomic shocks, which arise from di↵ering industry composition and the spread of COVID-
19. Columns (1) and (3) in Table 2 control only for week fixed e↵ects and Columns (2) and (4)
account for week-by-prefecture fixed e↵ects to address this concern. We are reassured that the
magnitude of our coe�cients are statically and economically robust. Our discussions will be based
on the results of week-by-prefecture fixed e↵ects.
5.3 Heterogeneous Impacts of Fiscal Payments
We explore heterogeneous responses in consumption to SFB payments by dividing account holders
into sub-samples. Table 2 and Figure 3 summarize the corresponding coe�cients and standard er-
rors of groups categorized by (a) family size, (b) COVID-19 income shocks, (c) liquidity constraints,
and (d) demand deposit balances and total asset holdings. In the Appendix, we report the results
for the sub-samples defined by quartiles of demand-deposit balance (Table B.2 and Figure B.1),
total wealth (Table B.3 and Figure B.2), and monthly salary (Table B.4 and Figure B.3), respec-
tively. Hereafter we base our discussion on the specifications of total outflows as the dependent
variable with control for prefecture-by-week fixed e↵ects.
(a) Family Size We first consider the heterogeneity in family size to check the validity of normal-
izing spending to a per-person amount. Family size is identified by the amount of SFB payments
because the payment per person is fixed at 100,000 Japanese Yen. Regression results for single-,
two-, three- and four-person families are shown in Column (5), (6), (7), and (8), respectively, in
Table 2. The coe�cients and standard errors are also plotted in Panel (a) of Figure 3. Larger
families tended to respond slightly more during the week of the deposit, though there is not much
di↵erence in expenditure per person.18
18Our bank account data do not have detailed information about family structure such as the number of children.
106
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 9
0-12
3
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
Figure 3: Heterogeneous Responses to SFB Payments
●
●
●
●
●
●
●●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●● ●
●
●
●
●
●●
●
●●
●
●
●
●
●
●
●
●
●
●
−20000
0
20000
−5 −4 −3 −2 −1 0 1 2 3 4 5Week relative to fiscal payment
Coe
ffici
ents
Amount of Payments (Family Size)●
●
●
●
100000 JPY
200000 JPY
300000 JPY
400000 JPY
(a) Amount of Payments or Family Size
● ● ●●
●
●
●
● ●
●
●
●●
●
●
●
●
● ●
●
●
●
●
●●
●
●
●
●
●●
●
●
−20000
0
20000
−5 −4 −3 −2 −1 0 1 2 3 4 5Week relative to fiscal payment
Change in monthly salary●
●
●
1: Decline less than 15%
2: Decline 15−50%
3: Decline more than 50%
(b) Covid Income Shock
● ● ●●
●
●
●
●
●
●●
●
●●
●
●
●
●
● ●
●
●
−20000
0
20000
−5 −4 −3 −2 −1 0 1 2 3 4 5Week relative to fiscal payment
Coe
ffici
ents
Liquidity Constraint●
●
YES
NO
(c) Liquidity Constraint
●● ●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
● ●
●
●●
●
●
●
●
●●
● ●●
●
●
●
●
●
●
●
●
●●
●
●
●
−20000
0
20000
−5 −4 −3 −2 −1 0 1 2 3 4 5Week relative to fiscal payment
Group●
●
●
●
Cash Lower, Wealth Lower
Cash Lower, Wealth Higher
Cash Higher, Wealth Lower
Cash Higher, Wealth Higher
(d) Cash and Wealth
Notes : The figure plots the estimated coe�cients �k for k 2 {�5, . . . ,�1, 0, 1, . . . , 5}. Note that��1 is normalized to 0. Bars indicate 95 percent confidence intervals. Standard errors are clusteredat prefectural level.
107
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 9
0-12
3
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
(b) Income Shock attributable to the COVID-19 Crisis We show how the income shock
attributable to COVID-19 a↵ects responses to SFB payments. Columns (9) and (10) in Table 2
presents results for account holders whose monthly income fell by less than 15% and between 15%
and 50%, respectively. Column (11) shows coe�cients of groups that experienced drops of 50% or
larger. Coe�cients and standard errors appear in Panel (b) of Figure 3. The most seriously a↵ected
group has a modestly higher MPC. However, di↵erences among the three groups are unclear. That
may be partly because the Japanese government’s COVID-19 fiscal package includes temporary
leave benefits, which likely stabilize jobs and benefit small businesses by compensating for significant
income loss among the self-employed.
(c) Liquidity Constraint Columns (12) and (13) in Table 3 and Panel (c) of Figure 3 show dif-
ferential responses to SFB payments by liquidity-constrained households and others, respectively.
We observe a higher spike among the former during the week of payments. Households not con-
strained by liquidity react moderately but sustain more spending during the second to fifth weeks
after SFB payments. These findings might be because liquidity-constrained households need cash
for daily living, whereas other households buy non-necessities without rushing. Standard errors for
the former group are small after the payment, showing that households facing liquidity constraints
react homogeneously to SFB payments.
(d) Wealthy Hand-to-mouth Another dimension of household heterogeneity is the joint distri-
bution of liquid and illiquid assets. Columns (14)–(17) of Table 3 and Panel (d) of Figure 3 display
the event study analysis by high/low total asset holdings and demand deposit balance. Immediate
responses during the first and second weeks after receiving payment were sharp among groups with
lower cash savings. The jump evident for the group with higher assets is also steep. That is, MPC
is substantial among the “wealthy hand-to-mouth.” Those households likely need cash for daily
108
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 9
0-12
3
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
transactions. However, MPC is almost irrelevant to total asset holdings. Our results accord with
the theoretical implications by Kaplan and Violante (2014) and Kaplan et al. (2018).
6 Conclusion
This study examines households’ responses to a large-scale and universal cash payment program
in response to the COVID-19 pandemic in Japan. We obtain causal estimates under a natural
experimental design created by randomized timings of cash transfers. Moreover, high-resolution
bank account data help to deliver precise and robust results. We find a sizable MPC and significant
heterogeneity in financial status.
Unlike past recessions characterized by macroeconomic demand/supply shocks, the current
COVID-19 crisis is characterized with heterogeneous sector-level shocks (del Rio-Chanona et al.,
2020), particularly concentrated in the service industry. That heterogeneity might amplify these
shocks through input-output links among industry networks (Baqaee and Farhi, 2020). House-
hold MPCs might also be biased toward non-services. Guerrieri et al. (2020) theoretically predict
that such a biased consumption pattern significantly reduces the multiplier e↵ects of fiscal policies.
Although service workers supposedly have higher MPCs, they earn less from consumer spending
stimulated by cash transfers. These secondary or higher-round e↵ects are crucial policy considera-
tions warranting further study. Multiplier e↵ects throughout economic networks could be discerned
by estimating MPCs by worker’s occupation/industry and by items consumed. In future research
we intend to identify worker information from their bank accounts and decompose expenditures
into categories by linking our dataset with credit card data.
109
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 9
0-12
3
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
References
A. Adams-Prassl, T. Boneva, M. Golin, and C. Rauh. Inequality in the impact of the coronavirus
shock: Evidence from real time surveys. CEPR Discussion Paper Series, (No. DP14665), 2020.
S. Agarwal, C. Liu, and N. S. Souleles. The reaction of consumer spending and debt to tax
rebates—evidence from consumer credit data. Journal of political Economy, 115(6):986–1019,
2007.
T. Alon, M. Doepke, J. Olmstead-Rumsey, and M. Tertilt. The impact of covid-19 on gender
equality. Covid Economics: Vetted and Real-Time Papers, 4:62–85, 2020.
M. Ando, C. Furukawa, D. Nakata, and K. Sumiya. Fiscal responses to the co vid-19 crisis in japan:
The first six months. National Tax Journal, 73(3):901–926D, 2020.
A. J. Auerbach and Y. Gorodnichenko. Fiscal multipliers in japan. Research in Economics, 71(3):
411–421, 2017.
S. R. Baker, R. A. Farrokhnia, S. Meyer, M. Pagel, and C. Yannelis. How Does Household Spending
Respond to an Epidemic? Consumption during the 2020 COVID-19 Pandemic. The Review of
Asset Pricing Studies, 10(4):834–862, 07 2020a.
S. R. Baker, R. A. Farrokhnia, S. Meyer, M. Pagel, and C. Yannelis. Income, liquidity, and the
consumption response to the 2020 economic stimulus payments. NBER Working Paper Series,
(No.27097), 2020b.
D. Baqaee and E. Farhi. Nonlinear production networks with an application to the covid-19 crisis.
NBER Working Paper Series, (No.27281), 2020.
M. Belot, S. Choi, E. Tripodi, E. van den Broek-Altenburg, J. C. Jamison, and N. W. Papageorge.
110
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 9
0-12
3
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
Unequal consequences of covid 19 across age and income: Representative evidence from six
countries. Covid Economics: Vetted and Real-Time Papers, (38), 2020.
R. Blundell, M. Costa Dias, R. Joyce, and X. Xu. Covid-19 and inequalities. Fiscal Studies, 41(2):
291–319, 2020.
D. Bounie, Y. Camara, E. Fize, J. Galbraith, C. Landais, C. Lavest, T. Pazem, B. Savatier, et al.
Consumption dynamics in the covid crisis: Real time insights from french transaction & bank
data. Covid Economics: Vetted and Real-Time Papers, (59), 2020.
R. A. Braun and D. Ikeda. Why cash transfers are good policy in the covid-19 pandemic. Federal
Reserve Bank of Atlanta’s Policy Hub, (NO.04-202), 2020.
C. Broda and J. A. Parker. The economic stimulus payments of 2008 and the aggregate demand
for consumption. Journal of Monetary Economics, 68:S20–S36, 2014.
V. M. Carvalho, S. Hansen, A. Ortiz, J. R. Garcia, T. Rodrigo, S. Rodriguez Mora, and P. Ruiz de
Aguirre. Tracking the covid-19 crisis with high-resolution transaction data. CEPR Discussion
Paper Series, (No. DP14642), 2020.
R. Chetty, J. N. Friedman, N. Hendren, and M. Stepner. How did covid-19 and stabilization policies
a↵ect spending and employment? a new real-time economic tracker based on private sector data.
NBER Working Paper Series, (No.27431), 2020.
O. Coibion, Y. Gorodnichenko, and M. Weber. The cost of the covid-19 crisis: Lockdowns, macroe-
conomic expectations, and consumer spending. Covid Economics: Vetted and Real-Time Papers,
(20), 2020a.
O. Coibion, Y. Gorodnichenko, and M. Weber. How did us consumers use their stimulus payments?
NBER Working Paper Series, (No.27693), 2020b.
111
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 9
0-12
3
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
N. Cox, P. Ganong, P. Noel, J. Vavra, A. Wong, D. Farrell, and F. Greig. Initial impacts of
the pandemic on consumer behavior: Evidence from linked income, spending, and savings data.
Brookings Papers on Economic Activity, 2020.
R. M. del Rio-Chanona, P. Mealy, A. Pichler, F. Lafond, and D. Farmer. Supply and demand
shocks in the covid-19 pandemic: An industry and occupation perspective. Covid Economics:
Vetted and Real-Time Papers, (6), 2020.
D. Farrell, P. Ganong, F. Greig, M. Liebeskind, P. Noel, and J. Vavra. Consumption e↵ects of
unemployment insurance during the covid-19 pandemic. Available at SSRN 3654274, 2020.
E. Forsythe, L. B. Kahn, F. Lange, and D. Wiczer. Labor demand in the time of covid-19: Evidence
from vacancy postings and ui claims. Journal of Public Economics, 189(104238), 2020.
H. Fujiki and M. Tanaka. How do we choose to pay using evolving retail payment technologies?
evidence from japan. Journal of the Japanese and International Economies, 49:85–99, 2018.
J. Garcıa-Montalvo and M. Reynal-Querol. Distributional e↵ects of covid-19 on spending: A first
look at the evidence from spain. Economic Working Paper Series, (No. 1740), 2020.
V. Guerrieri, G. Lorenzoni, L. Straub, and I. Werning. Macroeconomic implications of covid-19:
Can negative supply shocks cause demand shortages? NBER Working Paper Series, (No.26918),
2020.
S. Hacioglu, D. Kanzig, and P. Surico. The distributional impact of the pandemic. CEPR Discussion
Paper Series, (DP15101), 2020.
C.-T. Hsieh, S. Shimizutani, and M. Hori. Did japan’s shopping coupon program increase spending?
Journal of Public Economics, 94(7-8):523–529, 2010.
112
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 9
0-12
3
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
D. S. Johnson, J. A. Parker, and N. S. Souleles. Household expenditure and the income tax rebates
of 2001. American Economic Review, 96(5):1589–1610, 2006.
G. Kaplan and G. L. Violante. A model of the consumption response to fiscal stimulus payments.
Econometrica, 82(4):1199–1239, 2014.
G. Kaplan, B. Moll, and G. L. Violante. Monetary policy according to hank. American Economic
Review, 108(3):697–743, 2018.
E. Karger and A. Rajan. Heterogeneity in the marginal propensity to consume: Evidence from
covid-19 stimulus payments. FRB of Chicago Working Paper, 2020.
S. Kikuchi, S. Kitao, and M. Mikoshiba. Who su↵ers from the covid-19 shocks? labor market het-
erogeneity and welfare consequences in japan. Covid Economics: Vetted and Real-Time Papers,
(40), 2020.
M. J. Kim and S. Lee. Can stimulus checks boost an economy under covid-19? evidence from south
korea. IZA Discussion Papers, (No.13567), 2020.
M. Koga and K. Matsumura. Marginal propensity to consume and the housing choice. Bank of
Japan Working Paper Series, (No.20-E-3), 2020.
K. Li, N. Z. Foutz, Y. Cai, Y. Liang, and S. Gao. The impact of covid-19 lockdowns and stimulus
payments on spending of us lower-income consumers. Available at SSRN 3681629, 2020.
Q. Liu, Q. Shen, Z. Li, and S. Chen. Stimulating consumption at low budget–evidence from a
large-scale policy experiment amid the covid-19 pandemic. Available at SSRN 3626518, 2020.
K. Misra and P. Surico. Consumption, income changes, and heterogeneity: Evidence from two
fiscal stimulus programs. American Economic Journal: Macroeconomics, 6(4):84–106, 2014.
113
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 9
0-12
3
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
K. Misra, V. Singh, and Q. Zhang. Impact of the cares act stimulus payments on consumption.
Available at SSRN 3663493, 2020.
L. Montenovo, X. Jiang, F. L. Rojas, I. M. Schmutte, K. I. Simon, B. A. Weinberg, and C. Wing.
Determinants of disparities in covid-19 job losses. NBER Working Paper Series, (No.27132),
2020.
V. M. Nygaard, B. E. Sørensen, and F. Wang. Optimal allocation of the covid-19 stimulus checks.
CEPR Discussion Paper Series, (No. DP15283), 2020.
OECD. Supporting livelihoods during the covid-19 crisis: Closing the gaps in safety nets. 2020.
J. A. Parker, N. S. Souleles, D. S. Johnson, and R. McClelland. Consumer spending and the
economic stimulus payments of 2008. American Economic Review, 103(6):2530–53, 2013.
M. D. Shapiro and J. Slemrod. Consumer response to tax rebates. American Economic Review, 93
(1):381–396, 2003.
M. D. Shapiro and J. Slemrod. Did the 2008 tax rebates stimulate spending? American Economic
Review, 99(2):374–79, 2009.
A. Sheridan, A. L. Andersen, E. T. Hansen, and N. Johannesen. Social distancing laws cause only
small losses of economic activity during the covid-19 pandemic in scandinavia. Proceedings of
the National Academy of Sciences, 117(34):20468–20473, 2020.
S. Shimizutani. Consumer response to the 1998 tax cut: Is a temporary tax cut e↵ective? Journal
of the Japanese and International Economies, 20(2):269–287, 2006.
T. Watanabe. The responses of consumption and prices in japan to the covid-19 crisis and the
tohoku earthquake. CJEB Working Papers, (No. 373), 2020.
114
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 9
0-12
3
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
T. Watanabe and T. Yabu. Japan’s voluntary lockdown. Covid Economics: Vetted and Real-Time
Papers, (46), 2020.
115
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 9
0-12
3
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
Appendix
A Additional Discussion about Timing of SFB Payments
As discussed in Section 2.2, the timing of SFB payments was driven by the administrative delays
and can be regarded as unpredictable. This Appendix supports that claim by regressing the week
of SFB deposits against demographic variables and geographic indicators.
Specifically, the dependent variable is the week in which a bank account receives an SFB pay-
ment. Independent variables include age, a female dummy, wealth, cash saving, COVID-19 shock
indicators, and a dummy for liquidity constraint. We also add prefecture dummies and municipality
indicators in some specifications.19
Estimation results for the timing of payments appear in Table A.1. Columns (1) and (4) are
the specifications without geographic dummies, Columns (2) and (5) include prefecture dummies,
and columns (3) and (6) add municipality dummies. The R-squared values are 0.027 and 0.083,
respectively, in Columns (1) and (2). With municipality dummies, the R-squared reaches 0.29.
While geography predicts timings of payments to some extent, there remains substantial variation
in timing that is not explained by geography.
Turning to demographic variables, coe�cients are estimated precisely given the large sample
size. However, the magnitudes of these coe�cients are small and correlate weakly with the timing
of SFB deposits. For example, an account holder 10 years older than the average individual will
receive an SFB payment only 0.1 weeks earlier. As such, our analysis suggests that the account
holders’ region of residence drives the timing of SFB payments. Little statistical evidence suggests
households endogenously manipulate the timing of payments.
19The prefecture is Japan’s largest unit of local government. There are 47 prefectures in total. Municipalities arethe lower unit of local government in each prefecture. The total number of municipalities is 1,741 as of October 1,2018
116
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 9
0-12
3
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
Table A.1: Correlation between Timing of Payments and Demographic Variables
Dependent variable:
The week of deposit
(1) (2) (3) (4) (5) (6)
Age �0.013 �0.016 �0.015 �0.011 �0.016 �0.015(0.002) (0.003) (0.002) (0.001) (0.003) (0.002)
Female 0.344 0.282 0.258 0.332 0.255 0.243(0.030) (0.013) (0.013) (0.056) (0.029) (0.019)
Wealth 0.00001 0.00001 0.00001 �0.00001 �0.00000 �0.00000(0.00001) (0.00000) (0.00000) (0.00001) (0.00000) (0.00000)
Saving 0.0001 0.0001 0.0001 0.0001 0.0001 0.0001(0.00003) (0.00002) (0.00002) (0.00001) (0.00001) (0.00001)
Family size �0.257 �0.277 �0.272 �0.125 �0.146 �0.141(0.020) (0.014) (0.013) (0.020) (0.013) (0.011)
Monthly salary in 2019 0.002 0.001 0.001(0.0004) (0.0002) (0.0001)
COVID19 shock 1 �0.102 �0.099 �0.088(0.011) (0.012) (0.005)
COVID19 shock 2 �0.115 �0.118 �0.152(0.014) (0.020) (0.015)
Liquidity constraint �0.577 �0.576 �0.512(0.017) (0.013) (0.012)
Constant 28.033 27.417(0.130) (0.157)
Observations 2,798,149 2,798,149 2,798,149 1,194,378 1,194,378 1,194,378R2 0.027 0.083 0.286 0.026 0.092 0.306Prefecture FE Yes YesMunicipality FE Yes Yes
Notes : The dependent variable is the week in which a bank account receives the fiscal paymentper person. Independent variables include age, a female dummy, wealth, an SFB payment dummy,COVID-19 shock indicators, and a dummy for liquidity constraint. Columns (1) and (4) are speci-fications without geographic dummies. Columns (2) and (5) include prefecture dummies. Columns(3) and (6) add municipality dummies. The unit of wealth and amounts of SFB payments are10,000 JPY. Standard errors are in parenthesis and clustered at prefecture level.
117
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 9
0-12
3
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
B Additional Tables and Figures
Table B.2: Regression Results by Quartile of Savings in Demand-deposit Accounts
1st 2nd 3rd 4th5 weeks prior to payment �246 103 �1, 051 �12, 450
(760) (397) (955) (8, 063)4 weeks prior to payment �293 �976 122 4, 969
(711) (810) (598) (5, 869)3 weeks prior to payment 527 �1, 254 1, 014 �2, 862
(612) (570) (615) (4, 493)2 weeks prior to payment 987 �63 433 9, 130
(300) (431) (572) (4, 124)Week of Payment 36, 816 18, 587 12, 110 8, 022
(219) (626) (861) (2, 653)1 week after payment 16, 301 11, 256 7, 911 9, 804
(1, 089) (763) (1, 462) (3, 946)2 weeks after payment 5, 879 3, 513 1, 736 9, 912
(324) (296) (930) (4, 268)3 weeks after payment 3, 136 2, 719 3, 004 10, 585
(940) (724) (842) (3, 359)4 weeks after payment 3, 162 870 2, 185 4, 074
(278) (698) (788) (2, 888)5 weeks after payment 2, 041 265 2, 578 13, 653
(234) (1, 174) (895) (5, 674)Sample Size 24, 409, 070 24, 624, 880 24, 527, 545 24, 521, 210Week FEWeek*Prefecture FE Yes Yes Yes Yes
Notes : Standard errors are clustered at prefectural level.
118
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 9
0-12
3
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
Table B.3: Regression Results by Quartiles of Total Asset Holdings
1st 2nd 3rd 4th5 weeks prior to payment �1, 145 351 �181 �12, 798
(189) (564) (454) (7, 560)4 weeks prior to payment 44 �2, 003 �711 6, 323
(644) (972) (835) (5, 536)3 weeks prior to payment �844 �1, 547 2, 295 �2, 478
(198) (524) (683) (4, 454)2 weeks prior to payment �116 �210 468 10, 340
(253) (439) (438) (4, 493)Week of Payment 36, 248 17, 094 12, 558 9, 606
(259) (525) (593) (2, 741)1 week after payment 15, 099 10, 524 7, 131 12, 516
(465) (865) (669) (4, 467)2 weeks after payment 6, 418 3, 642 4, 263 6, 638
(172) (406) (1, 957) (3, 134)3 weeks after payment 3, 694 2, 891 2, 896 9, 958
(186) (452) (1, 422) (3, 317)4 weeks after payment 2, 063 1, 385 3, 514 3, 437
(412) (550) (413) (3, 168)5 weeks after payment 1, 043 923 2, 266 14, 288
(220) (517) (919) (6, 259)Sample Size 24, 462, 550 24, 540, 670 24, 540, 495 24, 540, 880Week FEWeek*Prefecture FE Yes Yes Yes Yes
Notes : Standard errors are clustered at prefectural level.
119
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 9
0-12
3
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
Table B.4: Regression Results by Quartile of Monthly Salary
1st 2nd 3rd 4th5 weeks prior to payment �246 103 �1, 051 �12, 450
(760) (397) (955) (8, 063)4 weeks prior to payment �293 �976 122 4, 969
(711) (810) (598) (5, 869)3 weeks prior to payment 527 �1, 254 1, 014 �2, 862
(612) (570) (615) (4, 493)2 weeks prior to payment 987 �63 433 9, 130
(300) (431) (572) (4, 124)Week of Payment 36, 816 18, 587 12, 110 8, 022
(219) (626) (861) (2, 653)1 week after payment 16, 301 11, 256 7, 911 9, 804
(1, 089) (763) (1, 462) (3, 946)2 weeks after payment 5, 879 3, 513 1, 736 9, 912
(324) (296) (930) (4, 268)3 weeks after payment 3, 136 2, 719 3, 004 10, 585
(940) (724) (842) (3, 359)4 weeks after payment 3, 162 870 2, 185 4, 074
(278) (698) (788) (2, 888)5 weeks after payment 2, 041 265 2, 578 13, 653
(234) (1, 174) (895) (5, 674)Sample Size 24, 409, 070 24, 624, 880 24, 527, 545 24, 521, 210Week FEWeek*Prefecture FE Yes Yes Yes Yes
Notes : Standard errors are clustered at prefectural level.
120
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 9
0-12
3
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
Figure B.1: Heterogeneous Responses to SFB Payments by Quartile of Savings in Demand-depositAccounts
3rd Quartile 4th Quartile
1st Quartile 2nd Quartile
−5 −4 −3 −2 −1 0 1 2 3 4 5 −5 −4 −3 −2 −1 0 1 2 3 4 5
−20000
0
20000
40000
−20000
0
20000
40000
Week relative to the timing of fiscal payment
Coe
ffici
ents
Notes : The figure plots estimated coe�cients of �k for k 2 {�5, . . . ,�1, 0, 1, . . . , 5}. Note that ��1
is normalized to 0. The bars indicate 95 percent confidence intervals. Standard errors are clusteredat prefectural level.
121
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 9
0-12
3
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
Figure B.2: Heterogeneous Responses to SFB Payments by Quartile of Total Asset Holdings
3rd Quartile 4th Quartile
1st Quartile 2nd Quartile
−5 −4 −3 −2 −1 0 1 2 3 4 5 −5 −4 −3 −2 −1 0 1 2 3 4 5
−20000
0
20000
−20000
0
20000
Week relative to the timing of fiscal payment
Coe
ffici
ents
Notes : The figure plots estimated coe�cients of �k for k 2 {�5, . . . ,�1, 0, 1, . . . , 5}. Note that ��1
is normalized to 0. The bars indicate 95 percent confidence intervals. Standard errors are clusteredat prefectural level.
122
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 9
0-12
3
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
Figure B.3: Heterogeneous Responses to SFB Payments by Quartile of Monthly Salary
Salary in 2019 = Q3 Salary in 2019 = Q4
Salary in 2019 = Q1 Salary in 2019 = Q2
−5 −4 −3 −2 −1 0 1 2 3 4 5 −5 −4 −3 −2 −1 0 1 2 3 4 5
−10000
0
10000
20000
30000
−10000
0
10000
20000
30000
Week relative to the timing of fiscal payment
Coe
ffici
ents
Notes : The figure plots estimated coe�cients of �k for k 2 {�5, . . . ,�1, 0, 1, . . . , 5}. Note that ��1
is normalized to 0. The bars indicate 95 percent confidence intervals. Standard errors are clusteredat prefectural level.
123
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 9
0-12
3
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
Covid Economics Issue 62, 18 December 2020
Copyright: Lauren Russell and Chuxuan Sun
The effect of mandatory child care center closures on women's labor market outcomes during the COVID-19 pandemic
Lauren Russell1 and Chuxuan Sun2
Date submitted: 3 December 2020; Date accepted: 7 December 2020
The COVID-19 pandemic has had a dramatic effect on women's labor market outcomes. We assess the effects of state-level policies that mandated the closure of child care centers or imposed class size restrictions using a triple-differences approach that exploits variation across states, across time, and across women who did and did not have young children who could have been affected. We find some evidence that both of these policies increase the unemployment rate of mothers of young children in the short term. In the long term, the effects of mandated closures on unemployment become even larger and persist even after states discontinue closures, consistent with a permanent child care supply side effect.
1 Associate Professor of Practice, Fels Institute of Government, University of Pennsylvania.2 Master's candidate, Fels Institute of Government, University of Pennsylvania.
124
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 12
4-15
4
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
1 Introduction
The economic downturn ushered in by the COVID-19 pandemic stands in stark contrast to
previous recessions because it has disproportionately a�ected women. Alon et al. (2020b)
show that for every recession between 1948 and 2009, men's unemployment rates have in-
creased more than women's or the e�ects have been relatively equal. The 2020 recession is
the �rst recession where the unemployment rate for women has risen signi�cantly more than
the unemployment rate for men.
Many have hypothesized that two primary factors are responsible for the dramatic e�ects
on women's employment rates in the US: the concentration of women in sectors and occu-
pations disproportionately impacted by the pandemic and changes in child care availability
(Alon et al., 2020b; Dingel et al., 2020; Collins et al., 2020). There are a prior reasons to
believe that changes in child care availability will disproportionately a�ect mothers. Alon
et al. (2020a) use time-use data to show that mothers spend more time on childcare than
fathers in two-parent households. They also point out using US Census Bureau data that
single mother households are much more common than single father households (Alon et al.,
2020a). Dingel et al. (2020) document that 32 percent of the US workforce has a child under
age 14 and 9.4 percent have a child under age 6. They conclude therefore that child care
center closures will a�ect women's employment much more than men's employment but do
not directly quantify the extent to which child care availability drives employment e�ects.
Estimating worker �xed e�ects models using US Current Population Survey data, Collins
et al. (2020) show that mothers with children aged 13 or younger reduced their work hours by
�ve times as much as father's between March and April 2020. However, it is not clear what
portion of this decline is due to di�erences in the type of occupations chosen by mothers and
fathers as opposed to child care responsibilities.
Heggeness (2020) provides some direct evidence on the e�ects of child care availability on
mother's labor market outcomes. Using a di�erences-in-di�erences approach, she estimates
e�ects of early public school closures and stay-at-home orders on women's unemployment,
labor market attachment, and hours worked. She �nds that mothers in early closure states
125
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 12
4-15
4
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
were signi�cantly more likely to have a job but not be working as a result of early shutdowns
but found no immediate impact on labor market detachment or unemployment. Because
the analysis focuses on parents of school age children and the e�ect of school closures, the
results do not shed light on loss of child care for parents of young children - those �ve and
under.
Compared to school age children, very young children require more intensive care (Drago,
2009). While school age children may be capable of completing some tasks independently
(such getting dressed, retrieving and eating a snack, or entertaining themselves), younger
children require around-the-clock supervision and attention. Loss of child care for very young
children during hours that would otherwise be used for paid work may have an even more
dramatic e�ect on mothers' labor supply outcomes than loss of public school for a school-age
child.
Prior to the pandemic, 24% of children aged 5 and younger received center-based care from
a day care center, preschool, prekindergarten or other early childhood program, and 60%
participated at least one weekly in some type of non-parental care arrangement including
home-based day cares or care arrangements with a relative (U.S. Department of Education,
2016). By mid-March and early April, 16 states had mandated the closure of child care
centers, potentially limiting the ability of parents to access child care. Another twenty
states imposed class size restrictions, typically allowing classes to contain no more than 10
children.
In this paper, we assess the e�ects of these mandatory child care center closures and
class size limits on mothers' labor supply outcomes, including unemployment, detachment
from the labor force, shares of women who are employed but not working, and actual hours
worked. In contrast to Heggeness (2020), we are able to estimate longer-term rather than
just immediate e�ects of closures. Speci�cally, we are able to track employment outcomes six
months after closures or class size restrictions were �rst implemented. Our triple-di�erences
approach exploits variation across states, time, and womens' motherhood status.
Ultimately, we �nd that state-level mandates that forced closure of child care centers or
126
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 12
4-15
4
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
imposed class size limits had important e�ects on unemployment rates of mothers of young
children aged 0 to 5. We estimate that in the short-term (the �rst one to three months
at the beginning of the pandemic when closures were actually in e�ect) closures increased
unemployment rates by 2.7 percentage points, on average, with this e�ect being marginally
statistically signi�cant at the 10% level. However, unemployment e�ects persisted and grew
larger in the months after the closures were rescinded. Post-closure, states that reopened
child care centers and shifted to class size restrictions had unemployment rates that were 4
percentage points higher than they would have been had closures never been implemented.
Similarly, we estimate that post-closure, the three states that reopened child care centers
without class size restrictions had unemployment rates were 6.6 percentage points higher
than they otherwise would have been.
We also �nd statistically signi�cant e�ects of class size restrictions. Our estimates in-
dicate that states that kept centers open but mandated smaller class sizes increased the
unemployment rates of mothers of young children by 2 percentage points. States that im-
plemented class size restrictions tended to keep these in place much longer than closures, so
we have less precision to estimate e�ects once class size restrictions were relaxed. The point
estimate is similar in magnitude to the e�ect when class size restrictions were in place, but
the con�dence interval is very wide and still includes zero.
Though we lack data to directly test how child care availability changed by state, it's
plausible that early �nancial pressures directly caused by mandated closure or class size
restrictions caused some centers to close their doors permanently. The Center for American
Progress has estimated that meeting pandemic-related state guidelines would increase oper-
ating expenses for child care providers by 47%, on average (Jessen-Howard and Workman,
2020b). Most of these increased costs would take the form of personnel costs to comply with
reduced class size requirements as well as increased sanitation costs (Jessen-Howard and
Workman, 2020a). During mandatory closures, some centers also tried to continue paying
sta� even when centers were closed, meaning that centers were trying to make payroll when
revenues were at best reduced, or at worst, nonexistent. Even if centers could meet budget
127
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 12
4-15
4
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
shortfalls for a month or two, it's unlikely they would be able to absorb such cost increases
in the long term, potentially leading to permanent closures and a contraction in the supply
of child care.
A survey by the National Association for the Education of Young Children found that
nationally, 18% of child care centers were closed in July 2020 as a result of the pandemic,
even though all states had o�cially allowed child care centers to reopen by that time, which
is consistent with this type of permanent supply side response (National Association for
the Education of Young Children, 2020). The survey also predicted that closures would
become more widespread in the months that followed. Forty percent of respondents said they
were certain that they would close permanently within the year without additional public
assistance (National Association for the Education of Young Children, 2020). Unfortunately,
such support has not materialized.
All of this evidence suggests that as the pandemic stretches on, the supply of child care has
became more constrained. Our evidence indicates that this has had the notable downstream
e�ect of increasing unemployment rates for women of young children.
2 Mandatory Child Care Center Closures
A prominent aspect of the COVID-19 crisis is that it has involved stay-at-home orders, some
of which forced the closure of child care facilities. In March-April 2020, 16 states issued
orders that forced child care businesses to close, though most included an exemption which
allowed centers to stay open if they served the children of essential workers.
The other 34 states (plus DC) allowed childcare businesses to stay open, according to Child
Care Aware of America, an organization that works with state and local childcare resource
and referral agencies (Quinton, 2020). However, among these 34 states, 20 imposed class
size limits designed to increase social distancing and reduce the risk of COVID transmission
without a classroom. Table 1 identi�es the states that ordered the closure of child care
businesses, states that allowed child care centers to remain open without class size limits,
128
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 12
4-15
4
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
and states that allowed child care centers to remain open but imposed class size limits. Even
though Alabama initially ordered child care centers to close, this closure remained in e�ect
only for one week between March 19, 2020 and March 27, 2020 at which point the state
allowed centers to reopen with a class size limit of 11. Therefore, in our analysis we classify
Alabama as a class size limit state rather than a mandated closure state. Note that most
states that mandated child care center closures in the early months of the pandemic later
transitioned to class size limits once centers were allowed to reopen.
Table 1: States with Mandatory Child Care Center Closures
State Policy Date Closures
E�ective
Date
Reopening
Allowed
Notes
Alabama Closed, with option to
provide emergency care
March 19, 2020
(modi�ed March
27, 2020 to allow
operation with 11
or fewer children
in each room)
March 27, 2020 When reopened, group sizes
limited to 11. By May 21, 2020, no
group size restrictions in place.
Alaska Option to remain open E�ective April 24, 2020 group sizes
limited to 10 children; by May 21,
2020, no group size restrictions in
place.
Arizona Option to remain open
Arkansas Option to remain open E�ective July 17, 2020, group sizes
limited to 10 people, including sta�
California Option to remain open E�ective August 25, 2020, cohorts
limited to no more than 14
children
Colorado Option to remain open Required to operate with groups of
10 or fewer from April 1, 2020 -
June 4, 2020
Connecticut Option to remain open E�ective March 16, 2020, limited
group sizes to no more than 14
children
Delaware Closed, with option to
provide emergency care
April 6, 2020 June 15, 2020 When reopened, group sizes
limited
District of
Columbia
Option to remain open
129
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 12
4-15
4
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
Florida Option to remain open E�ective March 24, 2020, all
gatherings of 10 or more people
prohibited; O�ce of Child Care
Regulation Guidance for Child
Care Providers identi�ed no
speci�c group size restrictions on
June 5, 2020
Georgia Option to remain open Starting March 23, 2020, centers
must limit class sizes to no more
than 10 (including sta�)
Hawaii Closed, with option to
provide emergency care
March 23, 2020 May 7, 2020 Starting on March 19, 2020, no
group could be larger than 10,
including the sta� person. By
June 9, 2020, no group size
restrictions in place.
Idaho Option to remain open
Illinois Closed, with option to
provide emergency care
March 20, 2020 May 29, 2020 When reopened, class size limits
imposed (varied depending on age
group)
Indiana Option to remain open E�ective March 20, 2020 - May 22,
2020, no more than 20 children can
reside in a classroom
Iowa Option to remain open
Kansas Option to remain open
Kentucky Closed, with option to
provide emergency care
March 20, 2020 June 8, 2020
(Family-based
care); June 15,
2020
(Center-based
care)
E�ective June 8, 2020, class sizes
limited to 10 children; class size
limits relaxed to 15 children for 2
years+ on September 1, 2020
Louisiana Option to remain open E�ective March 22, 2020, any
gatherings of 10 or more
prohibited; Child Care Guidelines
released on July 13, 2020 that
explicitly limited class sizes (varied
depending on age group)
Maine Option to remain open
Maryland Closed, with option to
provide emergency care
March 26, 2020 May 16, 2020 When centers reopened, no more
than 15 total persons per room
Massachusetts Closed, with option to
provide emergency care
March 23, 2020 E�ective June 8,
reopning plans to
be submitted for
reveiew
beginning June
15, 2020
Centers must apply to re-open;
forms �rst reviewed June 15, 2020;
class size limitations imposed upon
reopening until July 25, 2020
Michigan Closed, with option to
provide emergency care
March 23, 2020 June 1, 2020 When reopened, highly
recommended that group sizes be
kept to 10 or fewer children
Minnesota Option to remain open
Mississippi Option to remain open
Missouri Option to remain open
130
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 12
4-15
4
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
Montana Option to remain open E�ective April 1, 2020, class sizes
limited to 10 children
Nebraska Option to remain open E�ective March 18, 2020, class
sizes limited to 10 children per
class; e�ective May 4, 2020, some
child care facilities permitted to
have up to 15 children per class
Nevada Option to remain open
New Hampshire Option to remain open Centers must comply with health
and safety guidelines; limited class
sizes to 10 or fewer
New Jersey Closed, with option to
provide emergency care
March 25, 2020 June 15, 2020 E�ective June 18, 2020, group sizes
limited to 10 (not including sta�)
New Mexico Option to remain open E�ective March 23, 2020,
gatherings limited to 5 people;
e�ective August 14, 2020, class
sizes limited to 10-20 (depending
on age group)
New York Closed, with option to
provide emergency care
(NYC); Option to
remain open (NY State)
NYC: April 3,
2020
E�ective June 26, 2020, class sizes
limited to 15
North Carolina Closed, with option to
provide emergency care
March 25, 2020 May 8, 2020
North Dakota Option to remain open E�ective April 1, 2020, class sizes
limited to 9 children and one sta�
person. E�ective July 21, 2020,
classes limited to 15 people per
room (children plus sta�)
Ohio Closed, with option to
provide emergency care
March 26, 2020 May 31, 2020 When reopened, reduced group
sizes; e�ective August 9 returned
to normal group sizes
Oklahoma Option to remain open E�ective March 24, 2020 - June
17, 2020, group sizes limited to 10
Oregon Closed, with option to
provide emergency care
March 25, 2020 May 15, 2020 E�ective March 16, 2020, class
sizes limited to 10 children (8 for
children under 24 months old)
Pennsylvania Closed, with option to
provide emergency care
March 19, 2020 May 8, 2020 (24
northern
counties); May
15, 2020 (12
remaining
counties)
Rhode Island Closed, with option to
provide emergency care
March 29, 2020 June 1, 2020 When reopened, group sizes
limited to stable groups of 20
children; e�ective June 29, 2020,
programs could seek DHS approval
to increase group sizes
South Carolina Option to remain open
South Dakota Option to remain open
131
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 12
4-15
4
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
Tennessee Option to remain open E�ective March 15, 2020, �all e�ort
should be made to limit
congregation of children and class
sizes to 10 or less.�
Texas Option to remain open E�ective May 18, 2020, class sizes
limited (varied depending on age)
Utah Option to remain open E�ective March 25, 2020, class
sizes limited to 10; e�ective April
29, 2020, class sizes limited to 20
Vermont Closed, with option to
provide emergency care
March 17, 2020 June 1, 2020 E�ective September 1, 2020, class
size restrictions that vary by age
group
Virginia Option to remain open E�ective March 18, 2020, class
sizes limited to 10; e�ective June
2, 2020, class sizes restrictions
relaxed (new restrictions varied by
age group); e�ective September 25,
2020, no class size restrictions
Washington Option to remain open E�ective June 26, 2020, classes
could contain no more than 22
children and adults
West Virginia Closed, with option to
provide emergency care
March 25, 2020 May 11, 2020 E�ective May 9, 2020, class sizes
limited to 10 (including sta�)
Wisconsin Option to remain open
Wyoming Closed, with option to
provide emergency care
March 19, 2020 April 28, 2020 When reopened, class sizes limited
to 10 (including sta�); restrictions
updated on June 8, 2020 to keep
total persons to 10 or less in each
room; e�ective September 16,
2020, class size restrictions
removed
Notes: Most information comes from cross-referencing sources from the Hunt Institute (2020), the FoodIndustry Association (2020), and Child Care Aware of America (2020). State-speci�c news articles andgovernment orders were also consulted. For more details, see full data appendix.
Even in states that did not o�cially mandate stay-at-home orders or class size limits,
child care centers were deeply a�ected. Some centers voluntarily closed their doors due to
health concerns, and others voluntarily decreased class sizes in accordance with state recom-
mendations to allow for more social distancing. Some parents decided not to send children
to child care centers, even if centers were open in their area (Quinton, 2020). Therefore,
even the states where childcare businesses technically had the ability to operate as normal
during the early months of the pandemic, parents may have experienced decreased child care
access.
Between March 21 and April 2020, the Bipartisan Policy Center and Morning Consult
132
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 12
4-15
4
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
conducted a national survey of 800 parents with children under age 5. They found that 60%
of child care programs were fully closed (Bipartisan Policy Center, 2020). Unfortunately,
these aggregate data do not report data separately by state, so it is impossible to directly
compare the share of child care centers closed in states that mandated closure versus those
that did not during the earliest months of the pandemic. The aggregate statistics reported
by the the aforementioned National Association for the Education of Young Children survey,
which found that 18% of child care centers were closed in July 2020, suggests that some but
not all of these centers had reopened once states relaxed their closure policies in April, May,
and June.
Though many mandates to close child care centers were sometimes part of a more general
stay-at-home order, state-imposed child care center closures are not perfectly correlated with
other types of closures such as public school closures (Heggeness, 2020). Some states that
closed public schools explicitly allowed child care centers to remain open (Hunt Institute,
2020; Food Industry Association, 2020; Child Care Aware of America, 2020). In the analysis
that follows, we investigate the independent e�ect of mandatory child care center closure
policies and class size limit policies on the labor market outcomes of mothers of young
children.
3 Data Description
We use three data sources for our analysis: state-level information on child care center closure
policies, the Household Pulse Survey, and the Current Population Survey. Our data on child
care center closure policies, including dates of announcement/implementation and dates of
reopenings, come primarily from cross-checking three sources: Hunt Institute (2020); Food
Industry Association (2020); Child Care Aware of America (2020). However, we also consult
state-speci�c news articles and press releases to con�rm this information. The online data
appendix reports speci�c language from these orders and a complete list of sources for each
state.
133
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 12
4-15
4
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
The Household Pulse Survey, a survey launched in April 2020 speci�cally to shed light on
COVID-19 related issues, is administered by the US Census Bureau. The short 20-minute
survey consists of questions related to employment status, spending patterns, food security,
housing, physical and mental health, access to health care, and educational disruptions (US
Census Bureau, 2020; Centers for Disease Control and Prevention, 2020). The weekly survey
provides a �near real-time snapshot� of COVID-19 experiences because there is only an 8 day
lag between when respondents �ll out the questionnaire and when the results are reported
Centers for Disease Control and Prevention (2020). Although the survey has the advantage
of asking questions most relevant to e�ects of the COVID-19 pandemic, data were �rst
collected only after state-level mandates for child care center closures. Therefore, we are
unable to use the Pulse Survey data for our main triple-di�erences analysis. The data also
fail to identify speci�c ages of children for respondents, so we cannot isolate reporting to
parents of children aged 0 to 5, the population for whom child care is relevant.
Instead, we rely on the basic monthly �les from the Current Population Survey, a monthly
survey of about 60,000 households sponsored by the US Census Bureau and the US Bureau
of Labor Statistics (Flood, Sarah and King, Miriam and Rodgers, Renae and Ruggles, Steven
and Warren, J. Robert, 2020). Sampled households are in the survey for four consecutive
months, are out for eight months, and then return for another four consecutive months before
leaving the sample permanently. A new group of respondents starts in each calendar month
at the same time another group completes its rotation.
Our microdata correspond to January 2019 to September 2020, though most of our anal-
ysis focuses on September 2019 to September 2020. We limit the sample to people aged
18-64, inclusive, to focus analysis on the working-age population. We drop anyone living in
group quarters or working in the armed forces. We drop New York from our sample because
New York City had a child care center closure policy while the rest of the state did not, so it
is impossible to assign either treatment or control status to the state. We also drop any in-
dividuals whose reporting of age, sex, and race is inconsistent across the months where they
report data to the CPS. Our triple-di�erences analysis uses the subset of data corresponding
134
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 12
4-15
4
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
to women with children aged 0 to 5 and women without any children.
4 Aggregate E�ects of the Pandemic on Women's Employment and
the Importance of Child Care
Before presenting our analysis of the causal e�ects of state-level child care closure policies,
we begin by presenting descriptive statistics on women's unemployment and the reported
importance of child care access across all states during the pandemic period. Figure 1 uses
CPS data to show unemployment rates of men and women pre and post-pandemic. Prior
to the pandemic, unemployment rates of both men and women aged 18-64 hovered around
3-4%. Then unemployment rates increased dramatically between February 2020 and April
2020, peaking at 15.4% for women and 13.1% for men. Consistent with Alon et al. (2020b)'s
analysis, we �nd the increase is much larger for women � an 11.2 percentage point increase
� compared to 8.5 percentage points for men between February and April. Unemployment
rates for both men and women declined between April 2020 and September 2020 but still
remain at very high levels relative to the pre-pandemic period. As of September 2020, the
women's unemployment rate still exceeded the men's unemployment rate by 0.5 percentage
points.
135
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 12
4-15
4
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
Figure 1: Men and Women's Unemployment Pre and Post COVID-19
Source: IPUMS-CPS, University of Minnesota, www.ipums.org
Notes: Authors' tabulations. The sample consists of people aged 18-64 in the labor force.
4.1 Importance of Child Care Access
Next, we use the Pulse data to investigate how many women are reporting that child care
issues are a signi�cant driver of their unemployment (Figure 2). 1 For these �gures, we limit
our sample to parents with children aged 18 and under.
Throughout the data collection period of April 23 to September 28, a signi�cant number
of parents are reporting that they are not working and that this is due to child care issues.
The fraction of mothers reporting not working due to COVID-19 related child care issues
is signi�cantly higher than for fathers. For example, in the July 16-July 21 survey, 11% of
mothers versus only 3% of fathers were not working due to COVID-19 related child care
1For more analysis of these data, see Heggeness and Fields (2020).
136
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 12
4-15
4
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
issues.
Figure 2: Percent of Parents Not Working Due to COVID-19 Related Child Care Issues
Source: Household Pulse Survey Public Use File, United States Census Bureau, www.census.gov/programs-surveys/household-pulse-survey/
Notes: Figure displays the percent of parents aged 18-64 who have at least one child under 18 and report they arenot working due to COVID-19 related child care issues among all respondents to the Pulse Survey. Group quarterobservations are dropped, and composite weights are used.
We also extend previous descriptive work by investigating di�erences by characteristics
of these mothers. Figure 3 reports the percent of single and married mothers not working
who cited COVID-19 child care issues as the cause. In April and May, single mothers were
more likely than married mothers to report not working due to COVID-19 related child care
issues. By May and June, single and married mothers were reporting similar rates. By July,
August, and September married mothers were more likely than single mothers to report not
working due to COVID-19 related child care issues.
137
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 12
4-15
4
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
Figure 3: Percent of Mothers Not Working Due to COVID-19 Related Child Care Issues
Source: Household Pulse Survey Public Use File, United States Census Bureau, www.census.gov/programs-surveys/household-pulse-survey/
Notes: Sample includes mothers who have at least one child under 18. Group quarter observations are dropped, andcomposite weights are used.
We also investigated heterogeneity in child care issues as a driver of unemployment by
race/ethnicity. We �nd in Figure 4 that race/ethnicity is not a strong predictor of which
mothers report that they are not working due to child care issues.
138
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 12
4-15
4
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
Figure 4: Percent of Mothers Not Working Due to COVID-19 Related Child Care Issues
Source: Household Pulse Survey Public Use File, United States Census Bureau, www.census.gov/programs-surveys/household-pulse-survey/
Notes: Sample includes mothers who have at least one child under 18. Group quarter observations are dropped, andcomposite weights are used.
5 E�ects of Mandatory Child Care Center Closures
Though these descriptive statistics reveal that in the aggregate child care access is important
for mothers' labor supply, it is not known whether state-level child care closures or class size
restrictions, as opposed to voluntary closures of child care centers or loss of home-based care
provided by acquaintances, friends, or relatives, had an independent impact on mothers'
labor market outcomes.
5.1 Triple-Di�erences Empirical Strategy
To study the e�ects of state mandated child care center closures and class size restrictions
on the employment of women during the pandemic, we use a triple-di�erences strategy.2
2For a derivation of the triple-di�erences estimator and a complete discussion of its identifying assumptions, we refer readersto Olden and Møen (2020).
139
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 12
4-15
4
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
Our empirical strategy uses three dimensions of variation: cross-state variation in which
states implemented mandates, cross-time variation in when mandates were implemented,
and cross-worker variation in whether a woman had young children who would potentially
need child care.
One challenge in estimating the e�ect of child care center closures is the decision to
close all child care centers may not be quasi-random. While we �nd evidence that women's
employment was on parallel trends prior to the start of the pandemic for states that did
and did not implement closures, it is possible that states that mandated the closure of child
care centers were hit harder by the pandemic at the time the decision was made to close
child care centers. Thus, women's employment could decline more in these states for reasons
unrelated to child care availability. For example, prior work has shown that women tend
to be over-represented in sectors and occupations that were impacted most severely by the
pandemic (Alon et al., 2020a).
If these child care closure mandates are correlated with pandemic severity, a di�erences-
in-di�erences analysis may con�ate impacts of the pandemic on job availability with impacts
through child care availability. By including women without children in the analysis, we are
able to isolate the child care availability e�ect. We omit women of older children from the
analysis because these mothers also experienced changing family obligations as many schools
and universities were closed or switched to remote learning formats.
We start by estimating triple-di�erences event study models with leads and lags 6 months
before and 6 months after closure and class size restrictions implementation:
yipst = γst + θpt + µps +6∑
j=−6
βjClosurejpst +
6∑j=−6
∆jRestrictionjpst +Xipstδ + ωi + εiast (1)
In this regression equation, yipst is a labor market outcome for woman i in state s and
month t who either is or is not a parent (p) of a child aged 0-5. Recall that because we omit
parents of older children from the analysis sample, any observation that is not a parent of
140
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 12
4-15
4
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
a child aged 0-5 is a non-parent. We control for state-speci�c shocks that vary over time
γst and include interactions for parent and time e�ects θpt and parent and state e�ects µps.
The matrix Xipst includes a rich set of individual-level controls including age, marital status,
education, and occupation �xed e�ects and a control for whether there is another adult in
the household. The panel structure of the CPS also allows us to include person �xed e�ects
(ωi). We cluster standard errors at the state level.
The CPS survey is conducted on the 19th of each month and asks respondents questions
about the previous week. Because all of our closure and restriction policies were e�ective
after March 12, April 2020 is the �rst month where labor market outcomes in the CPS
could have been directly a�ected by these mandates. Accordingly, for our event studies, the
omitted month is March 2020, a month prior to when closures or restrictions could have �rst
impacted labor market outcomes.
It is important to keep in mind that closures were rescinded after one month in Hawaii,
North Carolina, West Virginia, and Wyoming, after two months in Illinois, Maryland, Michi-
gan, Ohio, Oregon, Pennsylvania, Rhode Island, and Vermont, and after three months in
Delaware, Kentucky, Massachusetts, and New Jersey. Therefore, no state in the +4 to
+6 months still had closures in e�ect, though we still plot these coe�cients to investigate
whether there were longer-term e�ects on labor market outcomes that persisted after policies
were relaxed.
To account for potentially di�erent e�ects in months where closures or class size restric-
tions were in e�ect vs. time periods where they had been relaxed, our triple-di�erences
regression takes the following form:
yipst = γst + θpt + µps + βClosureInEffectpst
+ΨClosureDiscontLimitImposedpst + ΛClosureDiscontNoLimitpst
+∆LimitInEffectpst + ΠLimitDiscontpst+
Xipstδ + ωi + εiast
(2)
Our set of �ve treatment indicators captures every possible treatment status in the post-
141
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 12
4-15
4
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
policy period. ClosureInEffectpst equals 1 if person i was a parent of a young child in state s
where child care center closures were mandated in month t. ClosureDiscontLimitImposedpst
equals 1 for post-closure months once centers were allowed to reopen if class size limits
were imposed at that time. ClosureDiscontNoLimitpst equals 1 in post-closure months
once the closure policy was discontinued if no class size limits were imposed. Similarly,
LimitInEffectpst equals 1 if person i was a parent of a young child in state s where child
care centers were subject to class size limits in month t. LimitDiscontpst equals 1 in months
after class size limits were discontinued.
The identifying assumption for our triple-di�erences estimator is that there is no contem-
poraneous shock that di�erentially a�ects the outcomes of the treatment group (mothers
with young children) compared to the control group (women without children) in the same
state-months as state-mandated child care center closures or child care class size limits.
5.2 Results
Figures 5 and 6 show the results of the event study speci�cation for four labor market out-
comes: labor force detachment, unemployment, being employed but not working, and actual
hours worked last week. The plots show evidence of parallel trends, lending credence to the
identifying assumption. There are no obvious e�ects of closures on labor force detachment,
being employed but not working, or actual hours worked last week. By contrast, there is an
obvious jump in unemployment after closures are implemented, and this e�ect persists and
is statistically signi�cant in the fourth month to sixth after the closure is implemented.
The results for class size limits in Figure 6 show a similar pattern. There are no dis-
cernible e�ects on labor force detachment, being employed but not working, or actual hours
worked last week, but there is a statistically signi�cant increase in unemployment at the time
class size limits go into e�ect. Unlike for closures, the negative employment e�ects seem to
dissipate over time and are not statistically signi�cant by three months after implementation.
142
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 12
4-15
4
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
Figure 5: Triple-Di�erences Event Studies for Child Care Center Closure Policies
Source: IPUMS-CPS, University of Minnesota, www.ipums.org
Notes: Results from estimating equation 1 as described in text and then plotting the coe�cients on the closure policytime relative to implementation indicators: βj .
143
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 12
4-15
4
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
Figure 6: Triple-Di�erences Event Studies for Class Size Limits
Source: IPUMS-CPS, University of Minnesota, www.ipums.org
Notes: Results from estimating equation 1 as described in text and then plotting the coe�cients on the class limitpolicy time relative to implementation indicators: ∆j .
144
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 12
4-15
4
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
Table 2 shows the triple-di�erences estimates. The �rst point estimate in column 2
indicates that closures increased unemployment rates of mothers with young children by 2.7
percentage points in months when a closure was actually in e�ect. (Note that these were
the early months of the pandemic.) The second point estimate indicates that in post-closure
months where closures were discontinued but class size limits were imposed (later months of
the pandemic), unemployment rates were 4.0 percentage points higher than they otherwise
would have been. The third point estimate indicates that in post-closure months where
closures were discontinued and no class size limits were imposed, unemployment rates were
6.6 percentage points higher than they would have otherwise been.
145
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 12
4-15
4
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
Table 2: E�ect of Mandated Child Care Center Closures and Class Size Limits on Women'sLabor Market Outcomes
Labor Market Outcome(1) (2) (3) (4)
Not In theLabor Force
Unemployed Employed ButNot Working
Actual HoursWorked Last
Week
Post Closure with 0.004 0.027* 0.001 -0.82Closure in E�ect (0.005) (0.016) (0.013) (0.44)
x Mother of Child 0-5
Post Closure with -0.000 0.040** -0.015 -0.38Closure Discontinued But Class Limits (0.006) (0.017) (0.014) (0.47)
x Mother of Child 0-5
Post Closure with -0.004 0.066*** 0.017 0.41Closure Discontinued & No Limits (0.005) (0.016) (0.023) (0.48)
x Mother of Child 0-5
Post Class Size Limits with -0.002 0.020** 0.004 0.18Limits in E�ect (0.004) (0.009) (0.012) (0.48)
x Mother of Child 0-5
Post Class Size Limits x -0.002 0.018 0.006 0.04Limits Discontinued x (0.006) (0.023) (0.013) (0.91)Mother of Child 0-5
Number of Individuals 92,956 68,250 64,671 63,194Number of Observations 275,896 191,204 179,010 170,519
Source: IPUMS-CPS, University of Minnesota, www.ipums.org
Notes: Results from estimation of equation (2) as described in the text. All regressions include occupation�xed e�ects, age �xed e�ects, marriage status �xed e�ects, control for at least one other adult in thehousehold, person �xed e�ects, and all the double interactions (state by month �xed e�ects, mother ofyoung child x month �xed e�ects, and state x mother of young child �xed e�ects). Standard errors areclustered at the state level.
The estimates also show an e�ect of class size limits on unemployment rates of mothers
of young children: +2.0 percentage points in months where limits were in e�ect with this
e�ect statistically signi�cant at the 5% level. There is no statistically signi�cant e�ect in
post-class limit months once limits were discontinued, though the con�dence interval cannot
rule out e�ects as large as during months where the limits were actually in place.
146
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 12
4-15
4
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
5.3 Robustness
Because mothers of very young infants may have taken maternity leave and been una�ected
by changes in child care center availability, we assessed the robustness of our results to
de�ning mothers of young children as those with children aged 1-5 rather than 0-5. Table 3
shows that our results are robust to this change in the young mothers de�nition, though the
e�ect of mandated closures in the early months when such policies were actually in e�ect is
similar in magnitude but no longer statistically signi�cant.
Table 3: Robustness Check Dropping Mothers of Infants
Labor Market Outcome(1) (2) (3) (4)
Not In theLabor Force
Unemployed Employed ButNot Working
Actual HoursWorked Last
Week
Post Closure with 0.004 0.024 0.012 -0.59Closure in E�ect (0.005) (0.017) (0.015) (0.45)
x Mother of Child 1-5
Post Closure with -0.002 0.047** -0.005 -0.18Closure Discontinued But Class Limits (0.006) (0.018) (0.021) (0.62)
x Mother of Child 1-5
Post Closure with -0.002 0.063*** 0.014 0.64Closure Discontinued & No Limits (0.006) (0.017) (0.020) (0.43)
x Mother of Child 1-5
Post Class Size Limits with -0.001 0.026*** -0.004 0.41Limits in E�ect (0.004) (0.010) (0.010) (0.45)
x Mother of Child 1-5
Post Class Size Limits x -0.002 0.013 -0.008 -0.29Limits Discontinued x (0.007) (0.023) (0.016) (1.03)Mother of Child 1-5
Number of Individuals 90,287 66,395 62,915 61,623Number of Observations 265,562 184,762 172,965 165,710
Source: IPUMS-CPS, University of Minnesota, www.ipums.org
Notes: Results from estimation of equation (2) as described in the text except parents (mothers of youngchildren) are de�ned as those with a child aged 1 to 5. All regressions include occupation �xed e�ects, age�xed e�ects, marriage status �xed e�ects, control for at least one other adult in the household, person �xede�ects, and all the double interactions (state by month �xed e�ects, mother of young child x month �xede�ects, and state x mother of young child �xed e�ects). Standard errors are clustered at the state level.
147
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 12
4-15
4
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
We would have liked to directly examine the number of women reporting that they are
unemployed because of child care issues, but the CPS does not ask a question with response
choices that would allow us to investigate this. The only reasons respondents can cite for
being unemployed include (1) looking for �rst jobs, (2) re-entering after an extended work
absence, (3) have left a job, (4) temporary job ended, (5) laid o�, or (6) left job for another
reason. None of these has a de�nitive link with child care issues. The Pulse survey is also
poorly suited to investigating whether mothers of young children in states with closures or
mandates were more likely to report being unemployed due to child care issues as the data
cannot be disaggregated to include only mothers with young children.
Instead, we take advantage of a child care question asked on the March 2020 Annual Social
and Economic Supplement (ASEC). Speci�cally, the question asked whether paid child care
was needed for each child in the household. We de�ne a mother has requiring paid child
care for a child aged 0-5 if there is any child in her household aged 0-5 for whom �paid child
care is needed.� We have 4,550 mothers with a child aged 0 to 5 who responded to both the
ASEC and appear in the March basic monthly �le. Among those mothers, 34% have at least
one child who needs child care which is consistent with estimates from the U.S. Department
of Education (2016).
A challenge of using this question for our analysis is that it is only asked once per year.
We impute whether a mother needs child care in other months where she participates in
the CPS panel by carrying this March response forward and backwards in time. Recall
that sampled households are in the survey for four consecutive months, so if this household
appeared in the CPS in February, March, April, and May, we use the March response and
assign that same value to this household (mother) in February, April, and May. Then, we
re-run our triple di�erences analysis, rede�ning the treatment group as mothers of children
aged 0-5 who expressed a need for paid child care. The control group is the same as before
- women without any children.
We would expect this analysis to be somewhat less informative than our preferred anal-
ysis previously presented. We are not able to look at e�ects of the closure and limitation
148
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 12
4-15
4
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
policies past June 2020 because we do not have any treatment group coverage in August
or September (more than four months after March). Moreover, though it is reasonable to
assume that if a mother required paid child care in March, she also required it in other
months, that assumption could be incorrect if there were changes in her outside options
(availability of informal child care arrangements). We also have less statistical power due to
smaller sample sizes. Nevertheless, if the results are truly driven by child care access, labor
market e�ects of child care policies should be somewhat larger when estimating the triple
di�erences speci�cation on this sample.
In fact, this is generally what we �nd in Table 4. Though we lose statistical signi�cance
of some estimates due to larger standard errors, the point estimates, especially for the e�ects
of class size limits, are somewhat larger than in the main speci�cation. In contrast to the
results before, we now �nd a statistically signi�cant e�ect on a woman being employed but
not working when class size limits are in place, and an increase in actual hours worked
last week when closures were discontinued but class size limits were imposed. The latter
e�ect may seem counterintuitive but is actually consistent with Heggeness (2020) who found
that for women who continued working, women in early closure states reported working
more weekly hours than those in late closure states, perhaps because these women were
compensating for reduced hours worked by other household members or because they faced
increased work loads due to changing operations during COVID.
149
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 12
4-15
4
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
Table 4: Robustness Check With Mothers Who Need Paid Child Care as of March 2020
Labor Market Outcome(1) (2) (3) (4)
Not In theLabor Force
Unemployed Employed ButNot Working
Actual HoursWorked Last
Week
Post Closure with 0.002 0.023 0.018 1.17Closure in E�ect (0.007) (0.030) (0.027) (1.13)
x Mother of Child 0-5
Post Closure with 0.049 0.039 -0.040 4.13***Closure Discontinued But Class Limits (0.037) (0.033) (0.027) (1.31)
x Mother of Child 0-5
Post Closure with 0.023 0.118* 0.012 -0.16Closure Discontinued & No Limits (0.024) (0.063) (0.022) (1.21)
x Mother of Child 0-5
Post Class Size Limits with 0.019* 0.044* 0.054*** 1.23Limits in E�ect (0.011) (0.024) (0.020) (0.76)
x Mother of Child 0-5
Post Class Size Limits x 0.007 0.062 -0.004 4.39*Limits Discontinued x (0.030) (0.039) (0.044) (2.34)Mother of Child 0-5
Number of Individuals 74,900 55,435 52,578 51,542Number of Observations 222,238 155,847 145,887 139,839
Source: IPUMS-CPS, University of Minnesota, www.ipums.org
Notes: Results from estimation of equation (2) as described in the text except parents (mothers of youngchildren) are de�ned as those with a child aged 0 to 5 who needed paid child care in March 2020. Allregressions include occupation �xed e�ects, age �xed e�ects, marriage status �xed e�ects, control for atleast one other adult in the household, person �xed e�ects, and all the double interactions (state by month�xed e�ects, mother of young child x month �xed e�ects, and state x mother of young child �xed e�ects).Standard errors are clustered at the state level.
6 Conclusion
In the aggregate, the COVID-19 pandemic has had a substantial e�ect on women's labor
supply outcomes, especially relative to men's. In this paper, we examine whether state-level
policies that forced the closure of child care centers or regulated class sizes speci�cally had
a discernible impact on labor supply outcomes for mothers of young children. We �nd that
these policies did, in fact, increase unemployment rates of mothers of young children in
150
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 12
4-15
4
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
these states. Unfortunately, the negative e�ects do not dissipate once states reopen child
care centers, consistent with permanent e�ects on child care supply in these states.
These results underscore the importance of access to reliable child care in promoting
equitable labor market outcomes for men and women. Especially at a time when 4.5 million
child care slots are at risk of disappearing, emergency funding and longer-term solutions to
support child care centers are desperately needed or mothers of young children may continue
to experience persistent and permanent employment losses in the future (Jessen-Howard and
Workman, 2020a).
151
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 12
4-15
4
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
References
Alon, T., Doepke, M., Olmstead-Rumsey, J., and Tertilt, M. (2020a). The impact of COVID-
19 on gender equality. Covid Economics: Vetted and Real-Time Papers, (4):62�85.
Alon, T., Doepke, M., Rumsey-Olmstead, J., and Tertilt, M. (2020b). This time it's di�erent:
The role of women's employment in a pandemic recession. NBER Working Paper 27660.
Bipartisan Policy Center (2020). Nationwide survey: Child care in the
time of coronavirus. Report. https://bipartisanpolicy.org/blog/
nationwide-survey-child-care-in-the-time-of-coronavirus/.
Centers for Disease Control and Prevention (2020). Novel COVID-19 survey takes nation's so-
cial, mental �Pulse�. https://www.cdc.gov/coronavirus/2019-ncov/communication/
accomplishments/pulse-survey.html.
Child Care Aware of America (2020). State by state resources. Accessed 27 August 2020.
https://www.childcareaware.org/resources/map/.
Collins, C., Landivar, L. C., Ruppanner, L., and Scarborough, W. J. (2020). COVID-19 and
the gender gap in work hours. Gender Work Organization, pages 1�12.
Dingel, J. I., Patterson, C., and Vavra, J. (2020). Childcare obligations will constrain many
workers when reopening the us economy. Becker Friedman Institute Working Paper No.
2020-46. https://bfi.uchicago.edu/wp-content/uploads/BFI_WP_202046.pdf.
Drago, R. (2009). The parenting of infants: A time-use study. Monthly Labor Review,
132(10):33�43.
Flood, Sarah and King, Miriam and Rodgers, Renae and Ruggles, Steven and Warren,
J. Robert (2020). Integrated Public Use Microdata Series, Current Population Survey:
Version 8.0 [dataset]. Minneapolis, MN: IPUMS. https://doi.org/10.18128/D030.V8.0.
Food Industry Association (2020). Coronavirus emergency state-by-state child care land-
scape. Accessed 27 August 2020. https://www.fmi.org/docs/default-source/
152
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 12
4-15
4
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
state-a�airs/covid19-state-by-state-child-care-landscape.pdf?sfvrsn=
461eb9e2_8.
Heggeness, M. L. (2020). Estimating the immediate impact of the COVID-19 shock on
parental attachment to the labor market and the double bind of mothers. Review of
Economics of the Household, Forthcoming.
Heggeness, M. L. and Fields, J. M. (2020). Working moms bear the
brunt of home schooling while working during COVID-19. US Cen-
sus Bureau. https://www.census.gov/library/stories/2020/08/
parents-juggle-work-and-child-care-during-pandemic.html.
Hunt Institute (2020). COVID-19 resources & policy considerations: Child care state actions.
Accessed 27 August 2020. http://www.hunt-institute.org/covid-19-resources/
state-child-care-actions-covid-19/.
Jessen-Howard, S. and Workman, S. (2020a). Coronavirus pandemic could lead to permanent
loss of nearly 4.5 million child care slots. Center for American Progress.
Jessen-Howard, S. and Workman, S. (2020b). The true cost of providing safe child care
during the coronavirus pandemic. Center for American Progress.
National Association for the Education of Young Children (2020). Holding on until help
comes: A survey reveals child care's �ght to survive. Report.
Olden, A. and Møen, J. (2020). The triple di�erence estimator. Discussion Papers 2020/1,
Norwegian School of Economics, Department of Business and Management Science.
Quinton, S. (2020). Will child care be there when states reopen? PEW Trusts Stateline
Article. https://www.pewtrusts.org/en/research-and-analysis/blogs/stateline/
2020/04/27/will-child-care-be-there-when-states-reopen.
US Census Bureau (2020). Measuring household experiences during the coronavirus (COVID-
19) pandemic. https://www.census.gov/data/experimental-data-products/
household-pulse-survey.html.
153
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 12
4-15
4
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
U.S. Department of Education, N. C. f. E. S. (2016). Early Childhood Program Participation
Survey of the 2016 National Household Education Surveys Program. (ECPP-NHES:2016).
https://nces.ed.gov/nhes/tables/ECPP_HoursPerWeek_Care.asp.
154
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 12
4-15
4
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
Covid Economics Issue 62, 18 December 2020
Copyright: Christina Boll and Till Nikolka
Rightly blamed the ‘bad guy’? Grandparental child care and Covid-191
Christina Boll2 and Till Nikolka3
Date submitted: 9 December 2020; Date accepted: 14 December 2020
This study explores the link between regular grandparental child care and Sars-CoV-2 infection rates at the level of German counties. In our analysis, we suggest that a region’s infection rates are shaped by region-, household- as well as individual-specific parameters. We extensively draw on the latter, exploring the inner- and outer-family mechanisms fueling individual contact frequency to test the potential role of regular grandparental child care in explaining overall infection rates. We combine aggregate survey data with local administrative data for German counties and find a positive correlation between the frequency of regular grandparental child care and local Sars-CoV-2 infection rates. However, statistical significance of this relationship breaks down as soon as potentially confounding factors, in particular the local Catholic population share, are controlled for. Our findings do not provide valid support for a significant role of grandparental child care in driving Sars-CoV-2 infections and rather suggest that the frequency of outer-family contacts driven by religious communities might be a more relevant channel in this context. Our results cast doubt on simplistic narratives postulating a link between intergenerational contacts and infection rates.
1 We thank Paul David Boll for editorial advice.2 German Youth Institute, University of Munich and University of Applied Labour Studies.3 German Youth Institute and CESifo.
155
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 15
5-17
5
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
Introduction
Since the beginning of the pandemic, many studies have analyzed the driving forces behind the spread of Sars-CoV-2. This includes the role of social contacts for the prevalence of Sars-CoV-2 and the disease Covid-19 caused by the virus. In this context, the links between contacts within the family, potentially increased risks of infection and Covid-19 mortality have been investigated e.g. by Aparicio and Grossbard 2020, Arpino et al. 2020, Balbo et al. 2020, and Bayer and Kuhn, 2020. In our analysis, we suggest that a region’s infection rates are mainly shaped by the two region-specific parameters infection path and spatial distance and the two individual-specific parameters vulnerability and contact frequency. We extensively draw on the latter, exploring the inner- and outer-family mechanisms fueling contact frequency to test the potential role of regular grandparental child care in explaining overall infection rates. We study these relationships for Germany combining aggregate survey data with local administrative data and find a positive correlation between the frequency of grandparental child care and local Sars-CoV-2 infection rates. However, statistical significance of this relationship breaks down as soon as potentially confounding factors, in particular the local Catholic population share, are controlled for. Our findings suggest that the frequency of outer-family contacts driven by religious communities might be a more relevant channel of Sars-CoV-2 infections than grandparental child care.
Due to substantially higher mortality rates of the older persons infected with Sars-CoV-2, early studies in 2020 already pointed at the vulnerability of certain regions due to their demographic characteristics (Kashnitsky and Aburto 2020) as well as the prevalence of intergenerational relations (Balbo et al. 2020). Aparicio and Grossbard (2020) present evidence that the frequency of intergenerational co-residence in US states is positively related to Covid-19 fatalities per capita. Similar results are found by Bayer and Kuhn (2020) in a cross-country analysis with 24 countries. However, Arpino et al. (2020) cannot confirm these findings. They provide a comprehensive analysis of aggregated data on intergenerational family relations from the SHARE (Survey of Health, Ageing and Retirement in Europe) survey linked with information on registered Sars-CoV-2 test data and case fatality rates as published by national health agencies in several European countries. They do not find a robust relationship between infections or case fatality rates and their key variables of interest on the family level, including the frequency of intergenerational contacts, the share of intergenerational households, and the prevalence of grandparental child care in a region or country.1
With data for Germany, a comprehensive analysis of the potential link between the extent of grandparental child care support and Sars-CoV-2 infections has not yet been conducted. The SHARE data used in Arpino et al. (2020), for example, do not allow for an analysis on the fine-grained local level because location information of respondents is only made available on the level of federal states. Moreover, previous studies have not convincingly investigated the role of potentially confounding factors when analyzing the correlation between grandparental child care support and Sars-CoV-2 infections. Our study aims to fill these gaps and contribute to the existing literature with an analysis for the case of Germany.
We study a potential relationship between grandparental child care support and Sars-CoV-2 infection rates in Germany combining survey data and registered infections at the level of German local administrative units (counties; German “Kreise”). We draw on a comprehensive register of Sars-CoV-2 infections registered by the local German health authorities (Gesundheitsämter) since the beginning of the pandemic and link these data with aggregated survey data on grandparental child care support at the 1 Dowd et al. (2020) argue that results from these aggregate level analyses should not be taken as evidence against a link between intergenerational relations and Sars-CoV-2 infection risks. However, this is not the scope of analysis in Arpino et al. (2020). Moreover, due to the lack of data on infections and the frequency/intensity of family contacts on the individual level, all studies to date rely on aggregate survey and/or administrative regional data.
156
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 15
5-17
5
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
local regional level. We also provide additional micro foundations for our analysis regarding intergenerational support based on rich individual level survey data.
Theoretical considerations and hypotheses
We restrict our attention to registered Sars-CoV-2 cases at the county level among those 60 years or older relative to the population of those aged 60 or older. The reason for the age restriction is that we are interested in whether child care responsibilities lead to higher risk of infections among the older, grandparent population. In our main analysis, we use March 23, 2020 as a reference point when counting all infections among those aged 60 years or older in the respective county.2 March 23 was the date when first official policy restrictions were announced in Germany as a response to the accelerating spread of Sars-CoV-2.
We theorize that beyond individual vulnerability in terms of health status, which we capture with the individual’s age group affiliation, three mechanisms might have shaped an individual’s exposure to the Sars-CoV-2 virus on March 23, 2020 and may therefore have driven the number of infections in the population aged 60 and over at the county level at that time (see Figure 1).
The first mechanism refers to the infection path, since due to the high infectiousness of the Sars-CoV-2 virus, the moment in time when the first infection is measured takes the respective county to a higher level of virus dissemination, measured by the increase in infection numbers (e.g. Frieden and Lee 2020, Bouffanais and Lim 2020). Second, the spatial distance of people, captured by the settlement structure, moderates the infection path such that (c.p.) densely populated areas will reinforce and sparsely populated areas will decrease the dissemination path of the virus. The mechanism behind settlement structure is that spatial distance to other human beings is more easily kept in more spacious areas (e.g. Rader et al. 2020). Third, and of utmost importance to this paper, the frequency of social contacts, which can be subdivided into intra-familial and extra-familial contacts, is decisive. Although this mechanism is mutually linked to settlement structure and infection path, we argue that it is by itself a function of individuals’ daily lives which are shaped by individual and household context-related characteristics as well as macro-level norms, institutions and economic activity. Concerning the extra-familial sphere, the job context as well as non-job activities should shape an individual’s interpersonal interactions. Inside the family, interactions refer to individuals of the same generation (intra-generational), as well as to upward intergenerational (the children generation which might be engaged in eldercare for their parents), and downward intergenerational (the grandchildren generation which might be subject to grandparental care) interactions. It is this last type of intra-family contacts we are interested in in this paper.
2 We also conducted our analysis using infection rates in the total population as dependent variable; those results are slightly weaker but qualitatively similar and not statistically significantly different from results in our main analysis. They are available upon request.
157
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 15
5-17
5
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
Figure 1. Illustration of causal mechanisms explaining infection rates at the county level.
Source: own illustration.
Since grandparental child care (GPCC) is only one source of social contacts, our hypotheses, which structure our empirical investigation, will focus on potential drivers of GPCC and, at the same time, on extra-familial sources of social contacts which might be confounding variables when analyzing the link between GPCC and the observed overall infection rates in the elderly population on the county level. We argue that grandparental child care (GPCC) should be shaped by the household context, institutional child care facilities available on the local level as a substitute for within-family care (H1), and norms, i.e. reciprocity norms and the value attributed to intergenerational relations (H2).
In more detail, we rely on the extensive margin of regular GPCC in our main analysis, exploiting the information on the use of regular child care. Starting with the household context (H1), we expect that working parents, especially working single mothers, should increase the likelihood of GPCC, since grandparents could compensate for scarce parental time resources in this case (Hank and Buber, 2009). Compared to small children under three for which substitutes for parental care are more difficult to find, the presence of a (youngest) child aged 3-5 should be related to more grandparental care (Hank and Buber, 2009). However, grandparental child care should become the less frequent the higher the number of children is (Jappens and van Bavel, 2012). Institutional child care usage at the individual level should decrease the need for grandparental care (Albertini and Kohli, 2013). Further, concerning norms, we expect that Catholic denomination is related to stronger intergenerational family ties and should therefore increase the propensity of grandparental child care (H2). A central foundation of Catholic social thought is the so-called subsidiary concept emphasizing the role of intra-family solidarity for social support (e.g. Gundlach 1964, Althammer 2013). Thus, we expect Catholic denomination to be positively associated with the prevalence of social norms fostering within-family support, e.g. GPCC.
MECHANISMS II: DRIVERS OF CONTACT FREQUENCY
MECHANISMS Iinfection pathcountylog days since 1st patient
contact frequencyindividual
spatial distancecounty: settlement structure
infection rate per population (age 60+) county level, 23.03.2020
exposureindividual: age
intra-familial extra-familial
jobindividual: employment status
non-job relatedcontactsindividuale.g. Carnival 02/2020,
church events,
institutional childcare
downwardindividual: regular grandparentalchildcare (GPCC)
upwardindividual: eldercare
intragenerationalindividual: siblings, cousins
norms/demographics
state: intergenerational relationspopulation share using regular GPCC
county: social networksreligiosity: population share of catholics
county: population compositionage: population share >60, population share <18
economic activity
county
log median income
household
context
householdparents‘ employment statusage of youngest childhousehold typenumber of childreninstitutional childcare usage
institutional
childcare
facilities
countycoverage rate <3coverage rate 3-5
TARGET VARIABLE
H5 H1
H3 H4
H6
H7
H8
H2
H9
158
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 15
5-17
5
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
Regarding infection rates at the county level, we postulate that the infection path in terms of (log) days since the first patient increases the number of infections since the virus has had more time to disseminate (H3). We further hypothesize that infection rates increase with population density, measured in four settlement types (H4). Moreover, we suggest that children enrolled in public child care as well as parents engaged in job-related social contacts could provide a source of infection for the (grand-)parents they interact with. Therefore, we use institutional child care coverage rates for children below 3 and 3-5, respectively (H5), as well as (log) median income as a proxy for economic activity (H6) as explanatory variables in the infections equation.3 We thereby rely on previous studies suggesting a positive linkage between economic activity and infections (e.g. Adda 2016, Rader et al. 2020) and institutional child care, respectively. The intuition behind income is that a higher economic added value is related to more job-related contacts. Remote work, especially work from home, has been shown to be more likely offered to and used by the highly educated workforce (Alipour et al. 2020) at the top of the earnings distribution. Since in March 2020 remote work was not as prevalent as it is today, this counter effect was probably not strong enough to outweigh the opposite (sales and revenues driven) positive linkage between income and infections.
Due to a higher risk exposure of elderly people who take care of their grandchildren we expect that regular GPCC should be associated with higher infection rates (H7). However, we doubt that GPCC is the true source of this phenomenon. Rather, we suggest that influential third variables drive the association between GPCC and infections at the county level. We thereby follow Arpino et al. (2020) who argue that a stronger focus on within-family ties and a correspondingly lower weight of extra-family ties could serve as a shield protecting the elderly against the virus. As pointed out by the authors, elderly people with close relationships to their children and grandchildren might rely less on social contacts outside the family which might potentially involve even bigger threats of infections. Second, strong intergenerational relationships might affect family life; members might be more careful regarding social interactions outside the family; in some of these cases grandparents might also live close to their children and grandchildren or even in the same house. Third, there is also evidence that family ties and interactions can have a positive effect on psychological wellbeing and health, decreasing the risk of an infection (see e.g. Cohen 2020). In sum, these arguments would motivate a negative association between regular GPCC and infection rates.
Beyond the indirect channel via GPCC, Catholic denomination might impact infections directly via a networks mechanism related to the private (non-job) sphere. Religious activities and other ritual activities do not occur unless a sufficient number of followers is reached at the local level. In particular, religious events, e.g. worships and religiously motivated celebrations with a high number of participants such as weddings, are suspected to drive infection numbers (Lee et al. 2020, Salvador et al. 2020). Moreover, though not directly associated with religiosity in contemporary Germany, German Rhineland regions, i.e. Rhineland-Palatinate and North Rhine-Westphalia where many Catholic people live, are well known for their Carnival processions which take place in February. Thus, via the regional bracket, counties with a high population share of Catholics might exhibit higher infection numbers via the non-job networks channel (H8). An extensive literature has studied the role of religiosity for extra-familial social networks (see e.g. for Germany: Traunmüller 2009).
Moreover, due to a higher contact frequency among younger people, a high population share of minors and a low share of elderly people (aged 60 or over) should increase infection rates. In sum, population composition by age should play a role, too (H9). As a further control variable we include the share of 3 Regarding institutional child care it is debated whether children in institutional care could also be a driver of infections in the population. Recent studies argue that infected but asymptomatic children are likely to be a source of further Sars-CoV-2 contagions; Hippich et al., 2020, Laxminarayan et al., 2020.
159
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 15
5-17
5
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
foreign nationals per county as well as a dummy variable for eastern German counties, which belonged to the former GDR.
Empirical analysis
Data and descriptive statistics
For our subsequent analysis in the light of the hypothesized mechanisms we combine different data sources. We draw on individual-level survey data from the 6th wave of KiBS (Kinderbetreuungsstudie) which is administered by the German Youth Institute (Alt et al. 2020, Aust et al. 2018). The data was collected in 2017 and involved 36.800 interviews conducted among reference persons (“Auskunftspersonen”) of children in the target population below the age of 15 living in 249 selected counties in Germany. We apply survey weights as described in Alt et al. (2020) correcting for non-response bias based on a two-stage weighting procedure. First, observations are weighted according to administrative statistics information on the distribution of children according to age groups in the residence state. Second, the weighting also accounts for non-response behavior of reference persons according to different institutional child-care arrangements. However, our main results are not sensitive to using the weighted or the unweighted sample in our analysis. The variable of interest in our analysis is the indicator whether grandparents provide regular child care to their grandchildren. KiBS contains information on whether grandparents provide regular child care support for a child. Alternative answer options in the survey questionnaire were: only irregular use, no use.4 If a respondent chose “regular grandparental child care support”, he/she was asked for how many hours this support was provided in a regular week. In our analysis, we use the survey information of whether a child receives regular grandparental child care as a binary variable at the individual level as well as in form of weighted respondent shares at the county level. For our micro-founded analyses, we draw on further individual-level information from the KiBS survey such as a child’s age, the household composition and labor force participation. Second, we use county level data on Sars-CoV-2 infections in Germany, as collected by the local health authorities (Gesundheitsämter) and published by the Robert-Koch Institute, RKI (2020). As central information from these data we use case counts per 100,000 inhabitants at the county level (at different points in time). In our main analysis we use infection register data from before the first policy restrictions were announced on March 23, 2020 to circumvent any differential effect of these restrictions which could be correlated with our variables of interest. In addition to the total recorded Sars-CoV-2 cases, the RKI has also published corresponding figures by age groups. From the RKI data, we additionally calculate the time that has passed since the first Sars-CoV-2 case record in a given county.
Third, we use administrative data on sociodemographic and economic characteristics at the county level as provided by the interactive database INKAR (Indikatoren und Karten zur Raum- und Stadtentwicklung) of the Bundesinstitut für Bau-, Stadt- und Raumforschung.
In the Appendix, we present some descriptive statistics for the key variables of our analysis (see Tables A1-A4 in the appendix for detailed information). Table A1 depicts the shares of GPCC aggregated from the KiBS data for the 16 federal states and four county types distinguishing rural and urban counties. The data shows that GPCC is less common in northern than in southern states in Germany and also less common in the so-called city-states, Hamburg, Berlin and Bremen. This holds for all regular GPCC as well as for GPCC that covers more than 7 hours per week. Table A2 reports these shares separately by children’s age groups (0-3 vs. 3-5 vs. 6-14) revealing that regular GPCC support is most frequent in the
4 See Appendix B1 for the exact wording of the question in the questionnaire.
160
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 15
5-17
5
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
age group 3-5. Table A2 additionally shows the share of intense regular GPCC (more than 7 hours per week) per child age group. However, we argue that not intensity but regularity is decisive for infection transmission on the micro level, and a significant share of families using grandparental child care is required to establish a related link between GPCC and infection rates on the macro (county) level. We therefore adhere to our sample specification of families with 0-14 year old children, focusing on regular GPCC in our main and intense GPCC in our sensitivity analysis.
As aforementioned, the use of institutional child care arrangements in a county might be an important factor in this context if grandparents step in in case no other child care possibilities are available. Column 4 in Table A1 shows that institutional child care coverage for children between 3 and 5 is generally very high (>89%) in all states and county types. In contrast, it is considerably lower for children below the age of three (column 3) The coverage in cities (Städtischer Kreis) tends to be lower than in rural areas and large metropolitan areas (Kreisfreie Großstadt), but the variation between different county types is not as large as between federal states. In general, coverage rates for under-threes are higher in Eastern compared to Western Germany.
Table A3 denotes total infection numbers and numbers per 1,000 inhabitants by German federal states for March 23, 2020. Additional to our focus group of elderly people aged 60 and over, figures for the total population are presented. Concerning the elderly, infections per 1,000 inhabitants vary between 24.2 (Mecklenburg-Western Pomerania) to 185.6 (Baden-Wuerttemberg). With respect to the total population, cases range from 81 in Bremen to 4,723 in Bavaria. Table A4 depicts total infection numbers and numbers per 1,000 inhabitants by German federal states for 30 September, 2020 which is used as an alternative reference point in our robustness checks.
Main regression analysis
Obviously, GPCC is only a small piece of the puzzle explaining overall infection rates in the target population of elderly people. Therefore, our empirical strategy consists of two strands. First, we build on the theoretical underpinnings of grandparental child care. This micro foundation will undergo an empirical test relying on the KiBS data outlined above. Via its aggregated form on the county level, the population share using regular GPCC, the outcome variable on the micro level will enter the infections equation on the macro (county) level as the second strand of our empirical design.
Starting with the micro foundation of regular GPCC, Table 1 presents the results of linear probability regressions with the weighted sample of 33,259 children in the target population of those aged 0 to 14 in Germany reached by the KiBS survey. The dependent variable is whether a child in the sample is regularly taken care of by a grandparent. Indeed, the results in Table 1 document that the population share of Catholic population on the county level is strongly positively associated with the probability that the grandparents are involved in child care support, confirming our hypothesis H2. Depending on the specification, this roughly means that a 10% increase in the Catholic population share in a county is associated with a 0.4 to 0.6 percentage point increase in the probability that the grandparents of a child are regularly involved in child care.
Moreover, these micro-level results reveal that children between the age of three and five are more frequently taken care of by their grandparents; children aged six to ten and particularly those aged ten to 14 are less likely to receive regular grandparental care compared to the reference group of children under the age of three. As to parents’ labor force participation status, the reference category is defined as couples pursuing a traditional male breadwinner model, i.e. the male partner works and the female partner is out of the labor force. Relative to this reference group, single mothers in as well as out of the
161
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 15
5-17
5
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
labor force are more likely to receive child care support from the grandparents of their children. For single fathers this is only the case if they are in the labor force (the group of single fathers in the sample contains only 60 observations). For couples that pursue a dual earner model and couples where only the female partner participates in the labor market it is also more likely that grandparents of their children provide regular child care support. This analysis shows that certain family types, e.g. single mothers and dual earner couples rely on support by grandparents relatively more often than traditional male breadwinner families. In case of an available and used institutional child care arrangement, grandparents are less likely to be involved in regular child care support. In sum, the household context variables meet our expectations (H1).
In column 2, we additionally include information on the county settlement structure (rural or urban), in column 3 we include federal state dummies. Our previous findings are robust to the inclusion of these additional variables.
Table 1. Analysis of micro-level characteristics associated with regular grandparental child care support.
Catholic population share 0.0619*** 0.0406*** 0.0528*** (0.0104) (0.0107) (0.0172) Child age 0-2 (reference category) - - - Child age 3-5 0.0314*** 0.0276*** 0.0276***
(0.00736) (0.00736) (0.00737) Child age 6-10 -0.00776 -0.0140** -0.0143**
(0.00661) (0.00664) (0.00665) Child age 10-14 -0.127*** -0.132*** -0.132***
(0.00675) (0.00677) (0.00677) Single mother, working 0.129*** 0.129*** 0.128***
(0.0168) (0.0168) (0.0168) Single mother, not working 0.0761*** 0.0714** 0.0705**
(0.0289) (0.0289) (0.0289) Single father, working 0.103* 0.105* 0.109**
(0.0537) (0.0536) (0.0536) Single father, not working -0.0381 -0.0158 -0.00985
(0.113) (0.113) (0.113) Couple, male breadwinner (reference category) - - -
Couple, female breadwinner 0.0316** 0.0340** 0.0342**
(0.0143) (0.0143) (0.0143) Couple, dual earner 0.131*** 0.131*** 0.130***
(0.00555) (0.00555) (0.00555) Couple, not working 0.00214 0.00439 0.00501
(0.0154) (0.0153) (0.0154) 1 child in household (reference category) - - - 2 children in household -0.0161*** -0.0164*** -0.0162***
(0.00546) (0.00545) (0.00546) 3+ children in household -0.0552*** -0.0547*** -0.0547***
(0.00641) (0.00640) (0.00641) Institutional child care attendance -0.0460*** -0.0417*** -0.0416*** (0.00503) (0.00505) (0.00508) Rural county – sparsely populated (ref.) - - -
162
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 15
5-17
5
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
Rural county 0.0176** 0.0164* (0.00850) (0.00885) Urban county 0.00386 0.00394 (0.00752) (0.00877) Urban county (single metropolitan area) -0.0330*** -0.0357*** (0.00757) (0.00848)
State fixed effects X Observations 33,259 33,259 33,259
Note: Weighted OLS regressions (Linear Probability Model), *** p<0.01, ** p<0.05, * p<0.1, standard errors in parentheses.
We now turn to the question whether the presented data reveal any relationship between regular grandparental child care support and higher rates of Sars-CoV-2 infections in a county.
Table 2 reports results of OLS regression estimations at the county level linking aggregated information on regular grandparental child care support for children aged under 15 from KiBS at the county level5 with regional administrative statistics and the infection rates as introduced above. Using the logarithm of cumulative infection rates in a county on March 23 as the dependent variable, we subsequently introduce various regional variables to test our hypotheses and account for potentially confounding factors in the correlation analysis. We present population-size-weighted estimates. Results are qualitatively similar when running the regressions without population weights.
Column 1 shows a statistically significant positive relationship between log Sars-CoV-2 infection rates and the percentage share of grandparental child care in a county, supporting H7. In this specification, we only additionally account for the days since the first registered Sars-CoV-2 case in a county (which are significantly positively associated to infection rates, supporting H3). These otherwise unconditional correlation results could be interpreted as confirming the hypothesis that more frequent contact of the old-age population is associated with higher risk of infection (e.g. Aparicio and Grossbard 2020). In columns 2-5 we introduce further variables to our empirical model. Counties with higher median income and metropolitan counties exhibit higher infection rates in the old-age population, providing support for our hypotheses H4 and H6. Child care coverage rates and the population shares of non-Germans and elderly people, respectively, lack significant associations with county-specific infection rates.
Most importantly, when including the share of Catholic population in column 5, we observe that this variable is highly positively correlated with our dependent variable (confirming H8) and that it absorbs the effect of grandparental child care.6 That is, H7 has to be discarded in this specification. This finding shows that there is no valid link between grandparental child care and infection rates on the county level as soon as further region-specific characteristics are controlled for. Therefore, regional variation in grandparental care cannot be blamed as a driver of regional variation in infection rates. Although affiliation to Catholic religion increases the likelihood of grandparental child care, the Catholic effect is
5 We only include counties with at least 30 individual-level observations. Results are qualitatively similar without minimum number of observations restriction or when restricting to at least 20 or 50 observations per county. Average number of underlying individual-level survey observations per county in the analysis presented in Table 2 is 184. As described above, we apply survey non-response weights as described in Alt et al. 2020. 6 When including state fixed effects in the specifications in Table 2 the coefficient for grandparental support also becomes insignificant in columns 1-4. This suggests that the positive relationship between grandparental support and Sars-CoV-2 infections in these columns is rather driven by the variation between states. The estimation results including fixed effects are available upon request.
163
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 15
5-17
5
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
obviously not limited to the family sphere but drives infection rates even if its within-family effect is accounted for. We motivate this out-of-family effect with social networks related to religious beliefs. Our regression results in Table 2 support this suggestion.
The population share of minors is significant in the second and even more so in the third specification. We do not find any effect of the population share of those aged 60 and over. That is, hypothesis H9 can only partially be confirmed and H5 cannot be confirmed at all on the basis of our data.
Table 2. Regressions explaining log registered infections 60+ per 100.000 (March 23, 2020) at the county level.
Log registered infections per 100,000. (60+) Log days since first case 1.478*** 0.929*** 0.916*** 0.947*** 0.914***
(0.168) (0.186) (0.183) (0.184) (0.174) Share of regular grandparental 1.658** 1.607** 1.531** 1.594** 0.481 child care (below age 15) (0.708) (0.683) (0.671) (0.682) (0.651) Log pop./km2 -0.0587 -0.123** -0.0544 -0.0903
(0.0506) (0.0587) (0.0710) (0.0712) Log median income 1.747*** 1.273** 1.442* 1.176*
(0.581) (0.622) (0.751) (0.708) East Germany -0.0903 0.0179 -0.0844 0.311
(0.149) (0.149) (0.220) (0.273) Urban county 0.362*** 0.311** 0.286**
(0.131) (0.137) (0.128) Population share age 60+ -1.842 -1.748 1.969
(2.606) (2.767) (2.798) Population share under 18 7.371 9.052* 10.38**
(4.698) (5.187) (5.005) Share institutional child care 0.00186 0.00246 (below age 3) (0.00716) (0.00943) Share institutional child care 0.0164 0.0164 (age 3-5) (0.0142) (0.0154) Foreigner share -1.551 0.478 (1.706) (1.637) Catholic population share 1.375*** (0.257) Observations (counties) 197 197 197 197 188 Note: Only counties with at least 30 individual observations in KiBS survey data, population-weighted coefficient estimation. *** p<0.01, ** p<0.05, * p<0.1, standard errors in parentheses.
In our next step, we build on the findings from Table 2 and particularly on the interpretation, that Catholic denomination fuels infection rates both via a GPCC and a social networks channel.
On the one hand, following from H1 which has been confirmed in our GPCC estimations, a higher share of Catholics in the population should increase GPCC, which in turn proved to increase infection risks in our infection regressions (Table 2, colums 1-4). Following this reasoning, we could expect a stronger/significant relationship between regular grandparental support and infection rates in these counties.
On the other hand, as discussed in the empirical strategy section, we suppose that the out-of-family channel requires a sufficient population share of Catholics to become effective: To celebrate religious events with non-family members and to establish religion-related rituals, a threshold of peers with the same attitudes is required. As aforementioned, due to the region commonality, we subsume Carnival
164
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 15
5-17
5
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
processions in this category. We cannot measure the non-job networks channel directly, but we suggest that in counties lacking a significant share of Catholics, the out-of-family mechanism should be relatively weaker than the within-family mechanism, attributing regular GPCC a higher role in this environment. Conversely, in counties exhibiting a significant share of Catholics, we expect the regular GPCC effect to be relatively weaker and the out-of-family channel to be relatively stronger. To disentangle the GPCC from the Catholic out-of-family channel and to verify which argumentation is supported by our data, our next task is to test H7 again, this time in two different social environments, distinguished by the prevalence of Catholic denomination. While this factor entered Table 2 as a ‘shift effect’, it is allowed to interact with all independent variables in the following regressions. To this end, we divide our sample in two subsamples with the first comprising counties with a population share of Catholics below 20% and the second with the remaining counties. We selected a threshold of 20% Catholic population share as this divides our sample roughly in half. Since Eastern Germany lacks representation in the first group, we restrict both subsamples to West German counties. The first group comprises counties with only minor shares of Catholics spans counties in Northern Germany. In 2011 for example, Schleswig-Holstein exhibited a Catholics share of 6.4%, while Lower Saxony stood at 18.3% (Frerk 2018a). The second group comprising counties with Catholic population shares of 20% and over concentrates in Southern and Western Germany. In 2015, the population share of Catholics stood at 39.3% in North Rhine-Westphalia, at 42.2% in Rhineland-Palatinate, at 59.8% in the Saarland, at 34.5% in Baden-Wuerttemberg and at 51.2% in Bavaria (Frerk 2018b). We use the model specification presented in column 4 of Table 2 and run our OLS regressions based on the two aforementioned subsamples.
Table 3 reports the results. As can easily be seen, regular GPCC is not significant in either of the two county groups. Apparently, neither in South-Western Germany nor in the rest of the country, regular GPCC was significantly related to infection rates in March 2020. Even in the counties where Catholics form a relevant population subgroup, regular GPCC does not significantly relate to infection rates among the elderly population. If anything, the data point to the importance of extra-familial ties: The regression coefficient for regular grandparental support is larger in those counties with a Catholic share of less than 20%. However, the coefficients lack significance.
A cautious interpretation of this finding would be that it further strengthens our interpretation derived from Table 2, column 5: The ‘bad guy’ role of GPCC is lost as soon as relevant third variables come on stage.
Table 3. Regressions explaining log registered infections per 100.000 inhabitants 60+ (March 23, 2020) at the county level (West-Germany).
Log registered infections per 100,000. (60+) Cath.pop. <20% Cath.pop. >=20% Log days since first case -0.0394 -0.104 1.185*** 1.098***
(0.474) (0.486) (0.229) (0.225) Share of regular grandparental 1.684 1.953 0.188 0.0981 child care (below age 15) (2.131) (2.182) (0.869) (0.843) Log pop./km2 -0.0685 -0.124 -0.0648 -0.0472
(0.195) (0.212) (0.106) (0.103) Log median income 3.332* 3.528* 1.023 1.128
(1.845) (1.881) (0.960) (0.932) Urban county 0.464 0.533* 0.0529 0.194
(0.294) (0.312) (0.201) (0.203) Population share age 60+ 7.084 5.723 -6.007 -2.544
165
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 15
5-17
5
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
(6.591) (6.921) (4.714) (4.769) Population share under 18 -0.609 -2.524 3.914 8.906
(13.05) (13.44) (6.468) (6.571) Share institutional child care 0.0330 0.0305 -0.0136 -0.0136 (below age 3) (0.0234) (0.0239) (0.0135) (0.0131) Share institutional child care -0.0396 -0.0291 0.0361 0.0423* (age 3-5) (0.0344) (0.0377) (0.0246) (0.0239) Foreigner share -4.019 -2.164 -2.908 -0.735 (4.974) (5.660) (2.474) (2.546) Catholic population share -2.505 1.222** (3.538) (0.480) Observations (counties) 40 40 97 97 Note: Only western German counties with at least 30 individual observations in KiBS survey data, population-weighted coefficient estimation. *** p<0.01, ** p<0.05, * p<0.1, standard errors in parentheses.
Sensitivity analysis
We conduct two robustness checks. The first refers to our dependent variable in the GPCC equation. Deviating from the standard for the main analysis defined above, we use the share of regular GPCC for children aged 0-14 that covers more than 7 hours per week to check whether grandparental child care intensity makes a difference for our results.7 See Table A5 in the appendix for the detailed results. Table A5 replicates the model specifications of Table 2, the only difference lies with the specification of the GPCC variable, using the intensive margin instead of the extensive margin of GPCC. The main take-away from the results is that, different from the extensive margin, regular GPCC is insignificant in all specifications when the intensive margin is used: The prevalence of regular grandparental child care of more than 7 hours a week does not significantly relate to a county’s infection rates among elderly people. This is intuitive since, as aforementioned, for virus dissemination, the frequency of contacts should play a higher role than intensity in terms of duration.
The second robustness check concerns the reference point in time regarding registered infections. Deviating from the standard defined above, we base our analysis on 30 September, 2020 and check the stability of our results against this change. For our regressions, we use the variable specifications denoted in Table 2. Table A6 in the appendix provides detailed regression results. Analogous to the results derived for March 23, 2020, the parameter of regular GPCC turns insignificant as soon as the population share with Catholic denomination is taken into account. Regarding the other independent variables, results resemble those of March, too: Metropolitan areas and a high population share of minors are positively associated with infection numbers. Different from March results, the population share of the non-German population is now positively related to infection rates, too. Differences to the results in our main analysis could be driven by the changed regional distribution of hot spots between March and September, 2020 (see Table A4 in the appendix): during that time, the virus had been disseminating further. By the end of September, parts of central and northern Germany (e.g. some Hessian regions and the city-state of Hamburg) exhibit rather high infection dynamics. By contrast, states in North-Eastern Germany (i.e. Mecklenburg-Vorpommern, Sachsen-Anhalt, Brandenburg and Schleswig-Holstein) still feature lower infection rates. These states are characterized by rather low foreign population shares (see Table A1), which might have driven the positive linkage of foreign population share to infection rates by the end of September.
7 We also conducted the regressions at the micro level with the dependent variable GPCC (more than 7h per week). The results are qualitatively similar to our main analysis and available upon request.
166
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 15
5-17
5
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
Conclusion
This study explores the role of regular grandparental child care (GPCC) for Sars-CoV-2 infection rates at the level of German counties. Our results show that a significant association between the two vanishes as soon as the Catholic population share is accounted for. Although Catholic religion increases the likelihood of grandparental child care in a family, our regressions of infection rates on a range of county-related covariates show that the Catholic effect drives infection rates even if the population share using GPCC effect is accounted for. We motivate this out-of-family channel with social networks related to religious beliefs. Our suggestion is tentatively confirmed by regressions based on two subsamples of counties, differing in their Catholic population share. Even in the subsample with a notable prevalence of Catholics, the GPCC parameter lacks significance. However, its effect size is lower, tentatively confirming our network hypothesis assigning out-of-family mechanisms a relatively higher role in these environments.
Our main result still holds when we use September 30, 2020 instead of March 23, 2020 as a point of reference for the infection rates: Here too, the significance of regular GPCC is lost as soon as the population share with Catholic denomination is accounted for. By contrast, drawing on the intensive instead of the extensive margin regarding GPCC does not yield any significant association between GPCC prevalence and infection rates in any of the specifications. This does not contradict our main finding but is intuitive since for virus dissemination, the frequency of contacts should play a higher role than their intensity in terms of duration.
In sum, our findings cast doubt on simplistic narratives postulating a link between intergenerational contacts and infection rates. At least our data does not provide valid support for a significant role of grandparental child care in driving infections. Rather, our findings support previous evidence highlighting the decisive role of third variables which have to be taken into account. Exemplified with Catholic denomination, we show that region-specific third variables may enforce the postulated (intra-family) mechanism but at the same time fuel out-of-family channels that limit or, in our case, even eliminate the statistical relevance of grandparental child care. In principle, a protective effect of intra-familial ties would also be possible, but our data is not suited to investigating this. Our data does not provide reliable support for a ‘bad guy’ role of grandparental child care.
167
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 15
5-17
5
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
References
Adda, J., “Economic Activity and the Spread of Viral Diseases: Evidence from High Frequency Data”, Quarterly Journal of Economics, 131(2), pp. 891-941, 2016. Albertini, M. and M. Kohli, „The generational contract in the family: An analysis of transfer regimes in Europe“, European Sociological Review, 29(4), pp. 828-840, 2013. Alipour, J. V., O. Falck, and S. Schüller, „Homeoffice während der Pandemie und die Implikationen für eine Zeit nach der Krise“, ifo Schnelldienst, 73(7), pp. 30-36. 2020. Alt, C., J. Anton, B. Gedon, S. Hubert, K. Hüsken, K. Lippert, and V. Schickle, „DJI-Kinderbetreuungsreport 2019. Inanspruchnahme und Bedarf aus Elternperspektive im Bundesländervergleich“ München: DJI, 2020. Althammer, J. (Hrsg.), „Caritas in veritate. Katholische Soziallehre im Zeitalter der Globalisierung“, Duncker & Humblot, Berlin, 2013. Aparicio, A. and S. Grossbard, “Intergenerational residence patterns and COVID-19 fatalities in the EU and the US”, IZA Discussion Paper No. 13452, 2020. Arpino, B., V. Bordone, and M. Pasqualini, “No clear association emerges between intergenerational relationships and COVID-19 fatality rates from macro-level analyses”, Proceedings of the National
Academy of Sciences (PNAS), 117(32), pp. 19116-19121, 2020. Aust, F., J. v. d. Burg, D. Hess, and B. Knecht, „Methodenbericht - Kinderbetreuungsstudie 2017 und Zusatzuntersuchung ‚Pilotstudie: Räumliche Mobilität von Familien und Kindern in Zeiten der Digitalisierung‘“, Bonn: infas, 2018. Balbo, N., F.C. Billari, and A. Melgrano, “The strength of family ties and COVID-19”, Contexts, American Sociological Association, https://contexts.org/blog/structural-shocks-and-extreme-exposures/#balbo, 2020. Bayer, C. and M. Kuhn, „Intergenerational ties and case fatality rates: a cross-country analysis“, IZA Discussion Paper No. 13114, 2020. Bouffanais, R., S. S. Lim, “Cities—Try to predict superspreading hotspots for COVID-19”, Nature 583, pp. 352–355, 2020. Cohen, S., „Psychosocial Vulnerabilities to Upper Respiratory Infectious Illness: Implications for Susceptibility to Coronavirus Disease 2019 (COVID-19)“, Perspectives on Psychological Science, pp. 1-14, 2020. Dowd, J.B., P. Block, V. Rotondi, and M. C. Mills, “Dangerous to claim “no clear association” between intergenerational relationships and COVID-19”, PNAS 117(42), 2020. Frerk, C., „Deutschland (2): Der evangelische Norden“, https://fowid.de/meldung/deutschland-2-evangelischer-norden, 2018a.
Frerk, C., „Deutschland (3): Der katholische Süden und Westen“, https://fowid.de/meldung/ deutschland-3-katholische-sueden-und-westen, 2018b.
Frieden, T.R., C. T. Lee, “Identifying and interrupting superspreading events—Implications for control of severe acute respiratory syndrome coronavirus 2”, Emerg. Infect. Dis. 26, 1059–1066, 2020. Gundlach, G, „Die Ordnung der menschlichen Gesellschaft.“ 2 Bde., Köln, 1964.
Hank, K. and I. Buber, „Grandparents caring for their grandchildren: Findings from the 2004 Survey of Health, Ageing, and Retirement in Europe“, Journal of Family Issues, 30(1), pp. 53-73, 2009. Hippich, M., L. Holthaus, R. Assfalg, J.M. Zapardiel Gonzalo, H. Kapfelsperger, M. Heigermoser, F. Haupt, D.A. Ewald, T.C. Welzhofer, B.A. Marcus, S. Heck, A. Koelln, J. Stock, F. Voss, M. Secchi, L. Piemonti, K. Rosa, U. Protzer, M. Boehmer, P. Achenbach, V. Lampasona, E. Bonifacio, and A.G. Ziegler, „Public health antibody screening indicates a six-fold higher Sars-CoV-2 exposure rate than reported cases in children“, Med, 2, pp. 1-15, 2020.
168
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 15
5-17
5
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
Jappens, M. and J. van Bavel, „Regional family norms and child care by grandparents in Europe“, Demographic Research, 27(4), pp. 84-120, 2012.
Indikatoren und Karten zur Raum- und Stadtentwicklung (INKAR). Ausgabe 2020. Hrsg.: Bundesinstitut für Bau-, Stadt- und Raumforschung (BBSR) im Bundesamt für Bauwesen und Raumordnung (BBR), Bonn, 2020.
Kashnitsky, I. and J.M. Aburto,”COVID-19 in unequally ageing European regions”, World
Development, 2020.
Laxminarayan, R., B. Wahl, S. R. Dudala, K. Gopal, S. Neelima, K. J. Reddy, J. Radhakrishman, and J.A. Lewnard, „Epidemiology and transmission dynamics of COVID-19 in two Indian states“, Science,
370(6517), pp. 691-697, 2020.
Lee, E. C., N. I. Wada, M. K. Grabowski, E. S. Gurley, and J. Lessler, „The engines of Sars-CoV-2 spread“, Science, 370(6515), pp. 406-407, 2020.
Rader, B., S. V. Scarpino, A. Nande, A. L. Hill, B. Adlam, R. C. Reiner, D. M. Pigott, B. Gutierrez, A. E. Zarebski, M. Shrestha, J. S. Brownstein, M. C. Castro, C. Dye, H. Tian, O. G. Pybus, and M. U. G. Kraemer, “Crowding and the shape of COVID-19 epidemics”, Nature Medicine, 2020.
Robert Koch-Institut (RKI), “Fallzahlen in Deutschland”, dl-de/by-2-0, 2020.
Salvador, C., Berg, M., Q. Yu, S. M. Alvaro, and S. Kitayama, „Relational mobility predicts faster spread of COVID-19: a 39-country study”, Psychological Science, Vol. 31(10), pp. 1236-1244, 2020.
Traunmüller, R., „Religion und Sozialintegration: Eine empirische Analyse der religiösen Grundlagen sozialen Kapitals (Abhandlungen).“ Berliner Journal für Soziologie, 19(3), pp. 435-468, 2009.
169
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 15
5-17
5
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
Appendix A
Table A1. Share of children receiving regular grandparental child care and institutional child care in per cent (columns 1-4), population share catholic in per cent (column 5), population share w. foreign
nationality in per cent (column 6) in 2017.
State
Share of regular
grandparental child care
(below age 15)
Share of regular
grandparental child care
>7hrs/week
Share institutional child care
children 0-3
Share institutional child care children
Share Catholic
population
Share foreign
population Baden-Wuerttemberg 16.9 6.5 27.8 92.2 36.3 15.1 Bavaria 16.6 7.2 26.5 89.7 53.8 12.7 Berlin 12.4 4.6 43.8 90.4 8.8 17.6 Brandenburg 14.1 5.5 55.5 92.3 3.3 4.4 Bremen 13.4 5.8 25.9 82.8 10.9 17.4 Hamburg 12.5 3.7 43.5 85.9 9.9 16.2 Hesse 16.2 7.7 29.2 90.5 24.3 15.7 Mecklenburg-Vorpommern 11.9 5.6 55.8 93.8 3.2 4.3 Lower Saxony 14.5 6.3 29.2 90.3 17.5 9.0 North Rhine-Westphalia 15.2 7.6 25.7 88.8 40.9 12.8 Rhineland-Palatinate 16.8 7.6 30.0 93.7 44.2 10.6 Saarland 16.9 8.7 27.7 91.2 61.9 10.7 Saxony 16.1 6.3 50.6 93.7 3.6 4.6 Saxony-Anhalt 16.6 8.2 57.0 91.7 3.4 4.7 Schleswig-Holstein 14.2 6.0 31.3 89.3 6.0 7.7 Thuringia 18.3 9.0 53.4 94.7 7.7 4.5 Rural county – sparsely populated 16.5 7.5 37.6 91.5 27.7 6.2 Rural county 18.5 8.7 34.3 91.1 31.2 7.4 Urban county 17.2 8.0 28.2 90.8 36.9 11.5 Urban county (single metropolitan area) 13.2 5.3 34.8 89.5 27.4 17.3 Total 15.6 6.9 32.6 90.6 30.1 11.7 Source column 1-2: KiBS Survey, wave 6, 33,259 observations (weighted, s. Alt et al. 2020), lower number of observations due to missing information for weight calculation; column 3-6: Data from INKAR, 2020. Reference year for all data: 2017.
170
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 15
5-17
5
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
Table A2. Share of children receiving regular grandparental child care in per cent in 2017
(by age group).
Share of regular grandparental child care
State age 0-3 age 0-3,
>7hrs/week age 3-5 age 3-5,
>7hrs/week age 6-14 age 6-14,
>7hrs/week Baden-Wuerttemberg 15.8 9.1 22.0 7.8 15.6 5.0 Bavaria 15.9 9.5 23.0 10.2 14.5 5.4 Berlin 11.6 3.9 15.8 6.0 11.4 4.3 Brandenburg 11.9 4.4 20.3 8.9 12.7 4.8 Bremen 14.1 7.8 13.5 5.1 13.0 5.1 Hamburg 11.4 4.3 17.5 5.6 11.0 2.8 Hesse 14.5 9.4 18.8 7.5 15.9 7.1 Mecklenburg-Vorpommern 8.4 3.5 13.2 6.4 12.7 6.1 Lower Saxony 16.6 9.2 17.8 7.4 12.7 4.9 North Rhine-Westphalia 19.3 12.5 17.1 7.9 13.0 5.6 Rhineland-Palatinate 16.1 8.3 22.4 11.3 15.1 6.2 Saarland 17.7 11.1 23.5 10.9 14.4 7.2 Saxony 12.9 4.8 21.2 7.8 15.5 6.1 Saxony-Anhalt 14.8 7.0 19.7 9.4 16.2 8.1 Schleswig-Holstein 15.7 7.7 19.6 7.8 11.9 4.7 Thuringia 15.7 6.0 21.6 11.1 18.1 9.6 Rural county – sparsely populated 17.0 10.1 20.5 9.7 15.3 6.3 Rural county 20.7 12.5 25.0 11.3 16.0 6.8 Urban county 17.3 10.6 22.3 9.2 15.6 6.8 Urban county (single metropolitan area) 13.7 7.0 16.4 6.7 11.5 3.7 Total 15.9 8.9 19.6 8.3 14.1 5.6 Source: KiBS Survey, wave 6, 33,259 observations (weighted, s. Alt et al. 2020), lower number of observations due to missing information for weight calculation.
171
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 15
5-17
5
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
Table A3. Sars-CoV-2 infections and associated deaths by federal state according to RKI registers from March 23, 2020.
State
Infections (total)
Infections (per 100,000
inhabitants)
Infections (age 60+)
Infections (per 100,000,
age 60+) Baden-Wuerttemberg 15,330 138.5 4,114 185.6 Bavaria 18,251 139.6 4723 178.2 Berlin 2,901 79.6 454 64.9 Brandenburg 1,055 42.0 266 44.0 Bremen 312 45.7 81 56.1 Hamburg 2,578 140.0 473 139.6 Hesse 3,576 57.1 811 63.1 Mecklenburg-Vorpommern 425 26.4 94 24.2 Lower Saxony 4,767 59.7 1,265 72.7 North Rhine-Westphalia 15,674 87.4 3,777 100.9 Rhineland-Palatinate 3,069 75.1 671 76.4 Saarland 946 95.5 224 96.6 Saxony 2,288 56.1 585 55.5 Saxony-Anhalt 779 35.3 220 38.3 Schleswig-Holstein 1,274 44.0 377 57.0 Thuringia 886 41.3 221 40.8 Total 74,111 89.0 18,356 107.8 Source: registered Sars-CoV-2 infections, RKI (2020), own calculation of infections per 100,000 population by age group based on INKAR, 2020.
Table A4. Sars-CoV-2 infections and associated deaths by federal state according to RKI registers from Sept. 30, 2020.
State
Infections (total)
Infections (per 100,000
inhabitants)
Infections (age 60+)
Baden-Wuerttemberg 49,203 444.5 570.6 Bavaria 67,761 518.2 615.4 Berlin 14,326 393.1 314.1 Brandenburg 4,251 169.2 181.0 Bremen 2,385 349.2 272.3 Hamburg 7,750 420.9 474.1 Hesse 18,788 299.8 291.8 Mecklenburg-Vorpommern 1,168 72.6 72.4 Lower Saxony 20,025 250.9 259.0 North Rhine-Westphalia 69,283 386.4 383.1 Rhineland-Palatinate 10,629 260.2 254.0 Saarland 3,296 332.8 454.3 Saxony 7,151 175.4 206.5 Saxony-Anhalt 2,613 118.3 118.4 Schleswig-Holstein 4,723 163.0 177.2 Thuringia 4,056 189.3 258.9 Total 287,408 339.4 383.3 Source: registered Sars-CoV-2 infections, RKI (2020), own calculation of infections per 100,000 population by age group based on INKAR, 2020.
172
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 15
5-17
5
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
Table A5. Regressions explaining log registered infections 60+ per 100.000 (March 23, 2020) at the county level.
Log registered infections per 100,000 (60+). Log days since first case 1.428*** 0.892*** 0.885*** 0.914*** 0.889***
(0.169) (0.187) (0.184) (0.186) (0.174) Share of regular grandparental 1.189 1.175 1.127 1.221 0.433 child care, below age 15, >7hrs/week (1.061) (1.031) (1.009) (1.020) (0.969) Log pop./km2 -0.0709 -0.135** -0.0668 -0.105
(0.0511) (0.0592) (0.0719) (0.0712) Log median income 1.792*** 1.339** 1.484* 1.161
(0.587) (0.628) (0.759) (0.709) East Germany -0.0909 0.0211 -0.0529 0.345
(0.151) (0.151) (0.222) (0.271) Urban county 0.373*** 0.318** 0.293**
(0.133) (0.138) (0.128) Population share age 60+ -1.509 -1.462 2.304
(2.632) (2.796) (2.807) Population share under 18 7.689 9.066* 10.34**
(4.744) (5.243) (5.009) Share institutional child care
0.000228 0.00172 (below age 3) (0.00719) (0.00938) Share institutional child care 0.0183 0.0168 (age 3-5) (0.0143) (0.0154) Foreigner share -1.448 0.705 (1.726) (1.635) Catholic population share 1.439*** (0.260) Observations (counties) 197 197 197 197 188 Notes: Minimum number of individual-level observations per county in KiBS survey: 30, regression estimates weighted based on population size. Standard errors in parentheses, *** p<0.01, ** p<0.05, * p<0.1.
Table A6. Regressions explaining log registered infections 60+ per 100.000 (Sept. 30, 2020) at the county level.
Log registered infections per 100,000 (60+). Log days since first case 1.223*** 0.556*** 0.538*** 0.486*** 0.477***
(0.169) (0.181) (0.176) (0.176) (0.163) Share of regular grandparental 1.597** 1.932*** 1.851*** 1.609** 0.570 child care, below age 15 (0.715) (0.664) (0.646) (0.651) (0.609) Log pop./km2 0.0567 0.00316 -0.0670 -0.126*
(0.0492) (0.0565) (0.0678) (0.0666) Log median income 1.132** 0.679 -0.0193 -0.421
(0.565) (0.599) (0.717) (0.662) East Germany -0.280* -0.154 -0.0346 0.338
(0.145) (0.143) (0.210) (0.255) Urban county 0.344*** 0.377*** 0.349***
(0.126) (0.130) (0.119) Population share age 60+ -1.316 -0.625 2.216
(2.510) (2.641) (2.616) Population share under 18 10.30** 9.030* 9.768**
(4.524) (4.952) (4.678)
173
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 15
5-17
5
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
Share institutional child care -0.00710 -0.00752
(below age 3) (0.00684) (0.00882) Share institutional child care 0.00466 0.00622 (age 3-5) (0.0136) (0.0144) Foreigner share 3.339** 5.649*** (1.629) (1.530) Catholic population share 1.158*** (0.240) Observations (counties) 197 197 197 197 188 Notes: Minimum number of individual-level observations per county in KiBS survey: 30, regression estimates weighted based on population size. Standard errors in parentheses, *** p<0.01, ** p<0.05, * p<0.1.
Table A7. Regressions explaining log registered infections per 100.000 inhabitants 60+ (Sept. 30, 2020) at the county level (West-Germany).
Log registered infections per 100,000 (60+). Cath.pop. <20% Cath.Pop. >=20% Log days since first case -0.214 -0.257 0.771*** 0.714***
(0.463) (0.478) (0.182) (0.181) Share of regular grandparental 0.740 0.917 0.520 0.461 child care, below age 15 (2.083) (2.144) (0.691) (0.679) Log pop./km2 -0.325* -0.362* 0.0185 0.0300
(0.190) (0.208) (0.0845) (0.0831) Log median income 1.395 1.525 -0.161 -0.0916
(1.803) (1.848) (0.764) (0.751) Urban county 0.642** 0.687** 0.0310 0.123
(0.287) (0.307) (0.160) (0.164) Population share age 60+ 5.415 4.518 -5.201 -2.935
(6.443) (6.798) (3.751) (3.841) Population share under 18 5.139 3.877 2.205 5.473
(12.76) (13.20) (5.148) (5.293) Share institutional child care 0.0199 0.0182 -0.0211* -0.0211* (below age 3) (0.0229) (0.0235) (0.0107) (0.0108) Share institutional child care -0.0257 -0.0188 0.0176 0.0217 (age 3-5) (0.0336) (0.0371) (0.0195) (0.0193) Foreigner share 6.279 7.502 0.0666 1.489 (4.862) (5.560) (1.969) (2.051) Catholic population share -1.651 0.800** (3.475) (0.387) Observations (counties) 40 40 97 97 Notes: Only western German counties with at least 30 individual observations in KiBS survey data, population-weighted coefficient estimation. *** p<0.01, ** p<0.05, * p<0.1, standard errors in parentheses.
174
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 15
5-17
5
COVID ECONOMICS VETTED AND REAL-TIME PAPERS
Appendix B
The exact wording of the question on grandparental child care support in KiBS wave 6 is as follows: In welchem Umfang wird Ihr Kind von den Großeltern betreut? (How often is your child looked after by his
or her grandparents?)
Regelmäßig, mit ___ Stunden in einer typischen Woche (regularly, with __ hours in a typical week) … ____
Nach Bedarf (as required)...................................................................................................................…… ____
Gar nicht (not at all)................................................................................................................................. ____
175
Covi
d Ec
onom
ics 6
2, 18
Dec
embe
r 202
0: 15
5-17
5